Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_nl.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | #include <linux/drbd.h> | |
28 | #include <linux/in.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/slab.h> | |
32 | #include <linux/connector.h> | |
33 | #include <linux/blkpg.h> | |
34 | #include <linux/cpumask.h> | |
35 | #include "drbd_int.h" | |
265be2d0 | 36 | #include "drbd_req.h" |
b411b363 PR |
37 | #include "drbd_wrappers.h" |
38 | #include <asm/unaligned.h> | |
39 | #include <linux/drbd_tag_magic.h> | |
40 | #include <linux/drbd_limits.h> | |
87f7be4c PR |
41 | #include <linux/compiler.h> |
42 | #include <linux/kthread.h> | |
b411b363 PR |
43 | |
44 | static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int); | |
45 | static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *); | |
46 | static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *); | |
47 | ||
48 | /* see get_sb_bdev and bd_claim */ | |
49 | static char *drbd_m_holder = "Hands off! this is DRBD's meta data device."; | |
50 | ||
51 | /* Generate the tag_list to struct functions */ | |
52 | #define NL_PACKET(name, number, fields) \ | |
53 | static int name ## _from_tags(struct drbd_conf *mdev, \ | |
54 | unsigned short *tags, struct name *arg) __attribute__ ((unused)); \ | |
55 | static int name ## _from_tags(struct drbd_conf *mdev, \ | |
56 | unsigned short *tags, struct name *arg) \ | |
57 | { \ | |
58 | int tag; \ | |
59 | int dlen; \ | |
60 | \ | |
61 | while ((tag = get_unaligned(tags++)) != TT_END) { \ | |
62 | dlen = get_unaligned(tags++); \ | |
63 | switch (tag_number(tag)) { \ | |
64 | fields \ | |
65 | default: \ | |
66 | if (tag & T_MANDATORY) { \ | |
67 | dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \ | |
68 | return 0; \ | |
69 | } \ | |
70 | } \ | |
71 | tags = (unsigned short *)((char *)tags + dlen); \ | |
72 | } \ | |
73 | return 1; \ | |
74 | } | |
75 | #define NL_INTEGER(pn, pr, member) \ | |
76 | case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \ | |
77 | arg->member = get_unaligned((int *)(tags)); \ | |
78 | break; | |
79 | #define NL_INT64(pn, pr, member) \ | |
80 | case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \ | |
81 | arg->member = get_unaligned((u64 *)(tags)); \ | |
82 | break; | |
83 | #define NL_BIT(pn, pr, member) \ | |
84 | case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \ | |
85 | arg->member = *(char *)(tags) ? 1 : 0; \ | |
86 | break; | |
87 | #define NL_STRING(pn, pr, member, len) \ | |
88 | case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \ | |
89 | if (dlen > len) { \ | |
90 | dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \ | |
91 | #member, dlen, (unsigned int)len); \ | |
92 | return 0; \ | |
93 | } \ | |
94 | arg->member ## _len = dlen; \ | |
95 | memcpy(arg->member, tags, min_t(size_t, dlen, len)); \ | |
96 | break; | |
1d273b92 | 97 | #include <linux/drbd_nl.h> |
b411b363 PR |
98 | |
99 | /* Generate the struct to tag_list functions */ | |
100 | #define NL_PACKET(name, number, fields) \ | |
101 | static unsigned short* \ | |
102 | name ## _to_tags(struct drbd_conf *mdev, \ | |
103 | struct name *arg, unsigned short *tags) __attribute__ ((unused)); \ | |
104 | static unsigned short* \ | |
105 | name ## _to_tags(struct drbd_conf *mdev, \ | |
106 | struct name *arg, unsigned short *tags) \ | |
107 | { \ | |
108 | fields \ | |
109 | return tags; \ | |
110 | } | |
111 | ||
112 | #define NL_INTEGER(pn, pr, member) \ | |
113 | put_unaligned(pn | pr | TT_INTEGER, tags++); \ | |
114 | put_unaligned(sizeof(int), tags++); \ | |
115 | put_unaligned(arg->member, (int *)tags); \ | |
116 | tags = (unsigned short *)((char *)tags+sizeof(int)); | |
117 | #define NL_INT64(pn, pr, member) \ | |
118 | put_unaligned(pn | pr | TT_INT64, tags++); \ | |
119 | put_unaligned(sizeof(u64), tags++); \ | |
120 | put_unaligned(arg->member, (u64 *)tags); \ | |
121 | tags = (unsigned short *)((char *)tags+sizeof(u64)); | |
122 | #define NL_BIT(pn, pr, member) \ | |
123 | put_unaligned(pn | pr | TT_BIT, tags++); \ | |
124 | put_unaligned(sizeof(char), tags++); \ | |
125 | *(char *)tags = arg->member; \ | |
126 | tags = (unsigned short *)((char *)tags+sizeof(char)); | |
127 | #define NL_STRING(pn, pr, member, len) \ | |
128 | put_unaligned(pn | pr | TT_STRING, tags++); \ | |
129 | put_unaligned(arg->member ## _len, tags++); \ | |
130 | memcpy(tags, arg->member, arg->member ## _len); \ | |
131 | tags = (unsigned short *)((char *)tags + arg->member ## _len); | |
1d273b92 | 132 | #include <linux/drbd_nl.h> |
b411b363 PR |
133 | |
134 | void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name); | |
135 | void drbd_nl_send_reply(struct cn_msg *, int); | |
136 | ||
137 | int drbd_khelper(struct drbd_conf *mdev, char *cmd) | |
138 | { | |
139 | char *envp[] = { "HOME=/", | |
140 | "TERM=linux", | |
141 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", | |
142 | NULL, /* Will be set to address family */ | |
143 | NULL, /* Will be set to address */ | |
144 | NULL }; | |
145 | ||
146 | char mb[12], af[20], ad[60], *afs; | |
147 | char *argv[] = {usermode_helper, cmd, mb, NULL }; | |
148 | int ret; | |
149 | ||
c2ba686f LE |
150 | if (current == mdev->worker.task) |
151 | set_bit(CALLBACK_PENDING, &mdev->flags); | |
152 | ||
b411b363 PR |
153 | snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev)); |
154 | ||
155 | if (get_net_conf(mdev)) { | |
156 | switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) { | |
157 | case AF_INET6: | |
158 | afs = "ipv6"; | |
159 | snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6", | |
160 | &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr); | |
161 | break; | |
162 | case AF_INET: | |
163 | afs = "ipv4"; | |
164 | snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4", | |
165 | &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr); | |
166 | break; | |
167 | default: | |
168 | afs = "ssocks"; | |
169 | snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4", | |
170 | &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr); | |
171 | } | |
172 | snprintf(af, 20, "DRBD_PEER_AF=%s", afs); | |
173 | envp[3]=af; | |
174 | envp[4]=ad; | |
175 | put_net_conf(mdev); | |
176 | } | |
177 | ||
1090c056 LE |
178 | /* The helper may take some time. |
179 | * write out any unsynced meta data changes now */ | |
180 | drbd_md_sync(mdev); | |
181 | ||
b411b363 PR |
182 | dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb); |
183 | ||
184 | drbd_bcast_ev_helper(mdev, cmd); | |
70834d30 | 185 | ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC); |
b411b363 PR |
186 | if (ret) |
187 | dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n", | |
188 | usermode_helper, cmd, mb, | |
189 | (ret >> 8) & 0xff, ret); | |
190 | else | |
191 | dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n", | |
192 | usermode_helper, cmd, mb, | |
193 | (ret >> 8) & 0xff, ret); | |
194 | ||
c2ba686f LE |
195 | if (current == mdev->worker.task) |
196 | clear_bit(CALLBACK_PENDING, &mdev->flags); | |
197 | ||
b411b363 PR |
198 | if (ret < 0) /* Ignore any ERRNOs we got. */ |
199 | ret = 0; | |
200 | ||
201 | return ret; | |
202 | } | |
203 | ||
204 | enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev) | |
205 | { | |
206 | char *ex_to_string; | |
207 | int r; | |
208 | enum drbd_disk_state nps; | |
209 | enum drbd_fencing_p fp; | |
210 | ||
211 | D_ASSERT(mdev->state.pdsk == D_UNKNOWN); | |
212 | ||
213 | if (get_ldev_if_state(mdev, D_CONSISTENT)) { | |
214 | fp = mdev->ldev->dc.fencing; | |
215 | put_ldev(mdev); | |
216 | } else { | |
217 | dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n"); | |
fb22c402 PR |
218 | nps = mdev->state.pdsk; |
219 | goto out; | |
b411b363 PR |
220 | } |
221 | ||
b411b363 PR |
222 | r = drbd_khelper(mdev, "fence-peer"); |
223 | ||
224 | switch ((r>>8) & 0xff) { | |
225 | case 3: /* peer is inconsistent */ | |
226 | ex_to_string = "peer is inconsistent or worse"; | |
227 | nps = D_INCONSISTENT; | |
228 | break; | |
229 | case 4: /* peer got outdated, or was already outdated */ | |
230 | ex_to_string = "peer was fenced"; | |
231 | nps = D_OUTDATED; | |
232 | break; | |
233 | case 5: /* peer was down */ | |
234 | if (mdev->state.disk == D_UP_TO_DATE) { | |
235 | /* we will(have) create(d) a new UUID anyways... */ | |
236 | ex_to_string = "peer is unreachable, assumed to be dead"; | |
237 | nps = D_OUTDATED; | |
238 | } else { | |
239 | ex_to_string = "peer unreachable, doing nothing since disk != UpToDate"; | |
240 | nps = mdev->state.pdsk; | |
241 | } | |
242 | break; | |
243 | case 6: /* Peer is primary, voluntarily outdate myself. | |
244 | * This is useful when an unconnected R_SECONDARY is asked to | |
245 | * become R_PRIMARY, but finds the other peer being active. */ | |
246 | ex_to_string = "peer is active"; | |
247 | dev_warn(DEV, "Peer is primary, outdating myself.\n"); | |
248 | nps = D_UNKNOWN; | |
249 | _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE); | |
250 | break; | |
251 | case 7: | |
252 | if (fp != FP_STONITH) | |
253 | dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n"); | |
254 | ex_to_string = "peer was stonithed"; | |
255 | nps = D_OUTDATED; | |
256 | break; | |
257 | default: | |
258 | /* The script is broken ... */ | |
259 | nps = D_UNKNOWN; | |
260 | dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff); | |
261 | return nps; | |
262 | } | |
263 | ||
264 | dev_info(DEV, "fence-peer helper returned %d (%s)\n", | |
265 | (r>>8) & 0xff, ex_to_string); | |
fb22c402 PR |
266 | |
267 | out: | |
268 | if (mdev->state.susp_fen && nps >= D_UNKNOWN) { | |
269 | /* The handler was not successful... unfreeze here, the | |
270 | state engine can not unfreeze... */ | |
271 | _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE); | |
272 | } | |
273 | ||
b411b363 PR |
274 | return nps; |
275 | } | |
276 | ||
87f7be4c PR |
277 | static int _try_outdate_peer_async(void *data) |
278 | { | |
279 | struct drbd_conf *mdev = (struct drbd_conf *)data; | |
280 | enum drbd_disk_state nps; | |
21423fa7 | 281 | union drbd_state ns; |
87f7be4c PR |
282 | |
283 | nps = drbd_try_outdate_peer(mdev); | |
21423fa7 PR |
284 | |
285 | /* Not using | |
286 | drbd_request_state(mdev, NS(pdsk, nps)); | |
99432fcc PR |
287 | here, because we might were able to re-establish the connection |
288 | in the meantime. This can only partially be solved in the state's | |
289 | engine is_valid_state() and is_valid_state_transition() | |
290 | functions. | |
291 | ||
292 | nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN. | |
293 | pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid, | |
294 | therefore we have to have the pre state change check here. | |
21423fa7 PR |
295 | */ |
296 | spin_lock_irq(&mdev->req_lock); | |
297 | ns = mdev->state; | |
197296ff | 298 | if (ns.conn < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &mdev->flags)) { |
21423fa7 PR |
299 | ns.pdsk = nps; |
300 | _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
301 | } | |
302 | spin_unlock_irq(&mdev->req_lock); | |
87f7be4c PR |
303 | |
304 | return 0; | |
305 | } | |
306 | ||
307 | void drbd_try_outdate_peer_async(struct drbd_conf *mdev) | |
308 | { | |
309 | struct task_struct *opa; | |
310 | ||
311 | opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev)); | |
312 | if (IS_ERR(opa)) | |
313 | dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n"); | |
314 | } | |
b411b363 | 315 | |
bf885f8a AG |
316 | enum drbd_state_rv |
317 | drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force) | |
b411b363 PR |
318 | { |
319 | const int max_tries = 4; | |
bf885f8a | 320 | enum drbd_state_rv rv = SS_UNKNOWN_ERROR; |
b411b363 PR |
321 | int try = 0; |
322 | int forced = 0; | |
323 | union drbd_state mask, val; | |
324 | enum drbd_disk_state nps; | |
325 | ||
326 | if (new_role == R_PRIMARY) | |
327 | request_ping(mdev); /* Detect a dead peer ASAP */ | |
328 | ||
329 | mutex_lock(&mdev->state_mutex); | |
330 | ||
331 | mask.i = 0; mask.role = R_MASK; | |
332 | val.i = 0; val.role = new_role; | |
333 | ||
334 | while (try++ < max_tries) { | |
bf885f8a | 335 | rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE); |
b411b363 PR |
336 | |
337 | /* in case we first succeeded to outdate, | |
338 | * but now suddenly could establish a connection */ | |
bf885f8a | 339 | if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) { |
b411b363 PR |
340 | val.pdsk = 0; |
341 | mask.pdsk = 0; | |
342 | continue; | |
343 | } | |
344 | ||
bf885f8a | 345 | if (rv == SS_NO_UP_TO_DATE_DISK && force && |
d10a33c6 PR |
346 | (mdev->state.disk < D_UP_TO_DATE && |
347 | mdev->state.disk >= D_INCONSISTENT)) { | |
b411b363 PR |
348 | mask.disk = D_MASK; |
349 | val.disk = D_UP_TO_DATE; | |
350 | forced = 1; | |
351 | continue; | |
352 | } | |
353 | ||
bf885f8a | 354 | if (rv == SS_NO_UP_TO_DATE_DISK && |
b411b363 PR |
355 | mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) { |
356 | D_ASSERT(mdev->state.pdsk == D_UNKNOWN); | |
357 | nps = drbd_try_outdate_peer(mdev); | |
358 | ||
359 | if (nps == D_OUTDATED || nps == D_INCONSISTENT) { | |
360 | val.disk = D_UP_TO_DATE; | |
361 | mask.disk = D_MASK; | |
362 | } | |
363 | ||
364 | val.pdsk = nps; | |
365 | mask.pdsk = D_MASK; | |
366 | ||
367 | continue; | |
368 | } | |
369 | ||
bf885f8a | 370 | if (rv == SS_NOTHING_TO_DO) |
b411b363 | 371 | goto fail; |
bf885f8a | 372 | if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) { |
b411b363 PR |
373 | nps = drbd_try_outdate_peer(mdev); |
374 | ||
375 | if (force && nps > D_OUTDATED) { | |
376 | dev_warn(DEV, "Forced into split brain situation!\n"); | |
377 | nps = D_OUTDATED; | |
378 | } | |
379 | ||
380 | mask.pdsk = D_MASK; | |
381 | val.pdsk = nps; | |
382 | ||
383 | continue; | |
384 | } | |
bf885f8a | 385 | if (rv == SS_TWO_PRIMARIES) { |
b411b363 PR |
386 | /* Maybe the peer is detected as dead very soon... |
387 | retry at most once more in this case. */ | |
20ee6390 | 388 | schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10); |
b411b363 PR |
389 | if (try < max_tries) |
390 | try = max_tries - 1; | |
391 | continue; | |
392 | } | |
bf885f8a AG |
393 | if (rv < SS_SUCCESS) { |
394 | rv = _drbd_request_state(mdev, mask, val, | |
b411b363 | 395 | CS_VERBOSE + CS_WAIT_COMPLETE); |
bf885f8a | 396 | if (rv < SS_SUCCESS) |
b411b363 PR |
397 | goto fail; |
398 | } | |
399 | break; | |
400 | } | |
401 | ||
bf885f8a | 402 | if (rv < SS_SUCCESS) |
b411b363 PR |
403 | goto fail; |
404 | ||
405 | if (forced) | |
406 | dev_warn(DEV, "Forced to consider local data as UpToDate!\n"); | |
407 | ||
408 | /* Wait until nothing is on the fly :) */ | |
409 | wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0); | |
410 | ||
411 | if (new_role == R_SECONDARY) { | |
81e84650 | 412 | set_disk_ro(mdev->vdisk, true); |
b411b363 PR |
413 | if (get_ldev(mdev)) { |
414 | mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; | |
415 | put_ldev(mdev); | |
416 | } | |
417 | } else { | |
418 | if (get_net_conf(mdev)) { | |
419 | mdev->net_conf->want_lose = 0; | |
420 | put_net_conf(mdev); | |
421 | } | |
81e84650 | 422 | set_disk_ro(mdev->vdisk, false); |
b411b363 PR |
423 | if (get_ldev(mdev)) { |
424 | if (((mdev->state.conn < C_CONNECTED || | |
425 | mdev->state.pdsk <= D_FAILED) | |
426 | && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced) | |
427 | drbd_uuid_new_current(mdev); | |
428 | ||
429 | mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1; | |
430 | put_ldev(mdev); | |
431 | } | |
432 | } | |
433 | ||
19f843aa LE |
434 | /* writeout of activity log covered areas of the bitmap |
435 | * to stable storage done in after state change already */ | |
b411b363 PR |
436 | |
437 | if (mdev->state.conn >= C_WF_REPORT_PARAMS) { | |
438 | /* if this was forced, we should consider sync */ | |
439 | if (forced) | |
440 | drbd_send_uuids(mdev); | |
f479ea06 | 441 | drbd_send_current_state(mdev); |
b411b363 PR |
442 | } |
443 | ||
444 | drbd_md_sync(mdev); | |
445 | ||
446 | kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); | |
447 | fail: | |
448 | mutex_unlock(&mdev->state_mutex); | |
bf885f8a | 449 | return rv; |
b411b363 PR |
450 | } |
451 | ||
ef50a3e3 LE |
452 | static struct drbd_conf *ensure_mdev(int minor, int create) |
453 | { | |
454 | struct drbd_conf *mdev; | |
455 | ||
456 | if (minor >= minor_count) | |
457 | return NULL; | |
458 | ||
459 | mdev = minor_to_mdev(minor); | |
460 | ||
461 | if (!mdev && create) { | |
462 | struct gendisk *disk = NULL; | |
463 | mdev = drbd_new_device(minor); | |
464 | ||
465 | spin_lock_irq(&drbd_pp_lock); | |
466 | if (minor_table[minor] == NULL) { | |
467 | minor_table[minor] = mdev; | |
468 | disk = mdev->vdisk; | |
469 | mdev = NULL; | |
470 | } /* else: we lost the race */ | |
471 | spin_unlock_irq(&drbd_pp_lock); | |
472 | ||
473 | if (disk) /* we won the race above */ | |
474 | /* in case we ever add a drbd_delete_device(), | |
475 | * don't forget the del_gendisk! */ | |
476 | add_disk(disk); | |
477 | else /* we lost the race above */ | |
478 | drbd_free_mdev(mdev); | |
479 | ||
480 | mdev = minor_to_mdev(minor); | |
481 | } | |
482 | ||
483 | return mdev; | |
484 | } | |
b411b363 PR |
485 | |
486 | static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
487 | struct drbd_nl_cfg_reply *reply) | |
488 | { | |
489 | struct primary primary_args; | |
490 | ||
491 | memset(&primary_args, 0, sizeof(struct primary)); | |
492 | if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) { | |
493 | reply->ret_code = ERR_MANDATORY_TAG; | |
494 | return 0; | |
495 | } | |
496 | ||
497 | reply->ret_code = | |
1f552430 | 498 | drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force); |
b411b363 PR |
499 | |
500 | return 0; | |
501 | } | |
502 | ||
503 | static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
504 | struct drbd_nl_cfg_reply *reply) | |
505 | { | |
506 | reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0); | |
507 | ||
508 | return 0; | |
509 | } | |
510 | ||
511 | /* initializes the md.*_offset members, so we are able to find | |
512 | * the on disk meta data */ | |
513 | static void drbd_md_set_sector_offsets(struct drbd_conf *mdev, | |
514 | struct drbd_backing_dev *bdev) | |
515 | { | |
516 | sector_t md_size_sect = 0; | |
517 | switch (bdev->dc.meta_dev_idx) { | |
518 | default: | |
519 | /* v07 style fixed size indexed meta data */ | |
520 | bdev->md.md_size_sect = MD_RESERVED_SECT; | |
521 | bdev->md.md_offset = drbd_md_ss__(mdev, bdev); | |
522 | bdev->md.al_offset = MD_AL_OFFSET; | |
523 | bdev->md.bm_offset = MD_BM_OFFSET; | |
524 | break; | |
525 | case DRBD_MD_INDEX_FLEX_EXT: | |
526 | /* just occupy the full device; unit: sectors */ | |
527 | bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev); | |
528 | bdev->md.md_offset = 0; | |
529 | bdev->md.al_offset = MD_AL_OFFSET; | |
530 | bdev->md.bm_offset = MD_BM_OFFSET; | |
531 | break; | |
532 | case DRBD_MD_INDEX_INTERNAL: | |
533 | case DRBD_MD_INDEX_FLEX_INT: | |
534 | bdev->md.md_offset = drbd_md_ss__(mdev, bdev); | |
535 | /* al size is still fixed */ | |
536 | bdev->md.al_offset = -MD_AL_MAX_SIZE; | |
537 | /* we need (slightly less than) ~ this much bitmap sectors: */ | |
538 | md_size_sect = drbd_get_capacity(bdev->backing_bdev); | |
539 | md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT); | |
540 | md_size_sect = BM_SECT_TO_EXT(md_size_sect); | |
541 | md_size_sect = ALIGN(md_size_sect, 8); | |
542 | ||
543 | /* plus the "drbd meta data super block", | |
544 | * and the activity log; */ | |
545 | md_size_sect += MD_BM_OFFSET; | |
546 | ||
547 | bdev->md.md_size_sect = md_size_sect; | |
548 | /* bitmap offset is adjusted by 'super' block size */ | |
549 | bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET; | |
550 | break; | |
551 | } | |
552 | } | |
553 | ||
4b0715f0 | 554 | /* input size is expected to be in KB */ |
b411b363 PR |
555 | char *ppsize(char *buf, unsigned long long size) |
556 | { | |
4b0715f0 LE |
557 | /* Needs 9 bytes at max including trailing NUL: |
558 | * -1ULL ==> "16384 EB" */ | |
b411b363 PR |
559 | static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' }; |
560 | int base = 0; | |
4b0715f0 | 561 | while (size >= 10000 && base < sizeof(units)-1) { |
b411b363 PR |
562 | /* shift + round */ |
563 | size = (size >> 10) + !!(size & (1<<9)); | |
564 | base++; | |
565 | } | |
4b0715f0 | 566 | sprintf(buf, "%u %cB", (unsigned)size, units[base]); |
b411b363 PR |
567 | |
568 | return buf; | |
569 | } | |
570 | ||
571 | /* there is still a theoretical deadlock when called from receiver | |
572 | * on an D_INCONSISTENT R_PRIMARY: | |
573 | * remote READ does inc_ap_bio, receiver would need to receive answer | |
574 | * packet from remote to dec_ap_bio again. | |
575 | * receiver receive_sizes(), comes here, | |
576 | * waits for ap_bio_cnt == 0. -> deadlock. | |
577 | * but this cannot happen, actually, because: | |
578 | * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable | |
579 | * (not connected, or bad/no disk on peer): | |
580 | * see drbd_fail_request_early, ap_bio_cnt is zero. | |
581 | * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET: | |
582 | * peer may not initiate a resize. | |
583 | */ | |
584 | void drbd_suspend_io(struct drbd_conf *mdev) | |
585 | { | |
586 | set_bit(SUSPEND_IO, &mdev->flags); | |
fb22c402 | 587 | if (is_susp(mdev->state)) |
265be2d0 | 588 | return; |
b411b363 PR |
589 | wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); |
590 | } | |
591 | ||
592 | void drbd_resume_io(struct drbd_conf *mdev) | |
593 | { | |
594 | clear_bit(SUSPEND_IO, &mdev->flags); | |
595 | wake_up(&mdev->misc_wait); | |
596 | } | |
597 | ||
598 | /** | |
599 | * drbd_determine_dev_size() - Sets the right device size obeying all constraints | |
600 | * @mdev: DRBD device. | |
601 | * | |
602 | * Returns 0 on success, negative return values indicate errors. | |
603 | * You should call drbd_md_sync() after calling this function. | |
604 | */ | |
24c4830c | 605 | enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local) |
b411b363 PR |
606 | { |
607 | sector_t prev_first_sect, prev_size; /* previous meta location */ | |
608 | sector_t la_size; | |
609 | sector_t size; | |
610 | char ppb[10]; | |
611 | ||
612 | int md_moved, la_size_changed; | |
613 | enum determine_dev_size rv = unchanged; | |
614 | ||
615 | /* race: | |
616 | * application request passes inc_ap_bio, | |
617 | * but then cannot get an AL-reference. | |
618 | * this function later may wait on ap_bio_cnt == 0. -> deadlock. | |
619 | * | |
620 | * to avoid that: | |
621 | * Suspend IO right here. | |
622 | * still lock the act_log to not trigger ASSERTs there. | |
623 | */ | |
624 | drbd_suspend_io(mdev); | |
625 | ||
626 | /* no wait necessary anymore, actually we could assert that */ | |
627 | wait_event(mdev->al_wait, lc_try_lock(mdev->act_log)); | |
628 | ||
629 | prev_first_sect = drbd_md_first_sector(mdev->ldev); | |
630 | prev_size = mdev->ldev->md.md_size_sect; | |
631 | la_size = mdev->ldev->md.la_size_sect; | |
632 | ||
633 | /* TODO: should only be some assert here, not (re)init... */ | |
634 | drbd_md_set_sector_offsets(mdev, mdev->ldev); | |
635 | ||
d845030f | 636 | size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED); |
b411b363 PR |
637 | |
638 | if (drbd_get_capacity(mdev->this_bdev) != size || | |
639 | drbd_bm_capacity(mdev) != size) { | |
640 | int err; | |
02d9a94b | 641 | err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC)); |
b411b363 PR |
642 | if (unlikely(err)) { |
643 | /* currently there is only one error: ENOMEM! */ | |
644 | size = drbd_bm_capacity(mdev)>>1; | |
645 | if (size == 0) { | |
646 | dev_err(DEV, "OUT OF MEMORY! " | |
647 | "Could not allocate bitmap!\n"); | |
648 | } else { | |
649 | dev_err(DEV, "BM resizing failed. " | |
650 | "Leaving size unchanged at size = %lu KB\n", | |
651 | (unsigned long)size); | |
652 | } | |
653 | rv = dev_size_error; | |
654 | } | |
655 | /* racy, see comments above. */ | |
656 | drbd_set_my_capacity(mdev, size); | |
657 | mdev->ldev->md.la_size_sect = size; | |
658 | dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1), | |
659 | (unsigned long long)size>>1); | |
660 | } | |
661 | if (rv == dev_size_error) | |
662 | goto out; | |
663 | ||
664 | la_size_changed = (la_size != mdev->ldev->md.la_size_sect); | |
665 | ||
666 | md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev) | |
667 | || prev_size != mdev->ldev->md.md_size_sect; | |
668 | ||
669 | if (la_size_changed || md_moved) { | |
24dccabb AG |
670 | int err; |
671 | ||
b411b363 PR |
672 | drbd_al_shrink(mdev); /* All extents inactive. */ |
673 | dev_info(DEV, "Writing the whole bitmap, %s\n", | |
674 | la_size_changed && md_moved ? "size changed and md moved" : | |
675 | la_size_changed ? "size changed" : "md moved"); | |
20ceb2b2 LE |
676 | /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */ |
677 | err = drbd_bitmap_io(mdev, &drbd_bm_write, | |
678 | "size changed", BM_LOCKED_MASK); | |
24dccabb AG |
679 | if (err) { |
680 | rv = dev_size_error; | |
681 | goto out; | |
682 | } | |
b411b363 PR |
683 | drbd_md_mark_dirty(mdev); |
684 | } | |
685 | ||
686 | if (size > la_size) | |
687 | rv = grew; | |
688 | if (size < la_size) | |
689 | rv = shrunk; | |
690 | out: | |
691 | lc_unlock(mdev->act_log); | |
692 | wake_up(&mdev->al_wait); | |
693 | drbd_resume_io(mdev); | |
694 | ||
695 | return rv; | |
696 | } | |
697 | ||
698 | sector_t | |
a393db6f | 699 | drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space) |
b411b363 PR |
700 | { |
701 | sector_t p_size = mdev->p_size; /* partner's disk size. */ | |
702 | sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */ | |
703 | sector_t m_size; /* my size */ | |
704 | sector_t u_size = bdev->dc.disk_size; /* size requested by user. */ | |
705 | sector_t size = 0; | |
706 | ||
707 | m_size = drbd_get_max_capacity(bdev); | |
708 | ||
a393db6f PR |
709 | if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) { |
710 | dev_warn(DEV, "Resize while not connected was forced by the user!\n"); | |
711 | p_size = m_size; | |
712 | } | |
713 | ||
b411b363 PR |
714 | if (p_size && m_size) { |
715 | size = min_t(sector_t, p_size, m_size); | |
716 | } else { | |
717 | if (la_size) { | |
718 | size = la_size; | |
719 | if (m_size && m_size < size) | |
720 | size = m_size; | |
721 | if (p_size && p_size < size) | |
722 | size = p_size; | |
723 | } else { | |
724 | if (m_size) | |
725 | size = m_size; | |
726 | if (p_size) | |
727 | size = p_size; | |
728 | } | |
729 | } | |
730 | ||
731 | if (size == 0) | |
732 | dev_err(DEV, "Both nodes diskless!\n"); | |
733 | ||
734 | if (u_size) { | |
735 | if (u_size > size) | |
736 | dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n", | |
737 | (unsigned long)u_size>>1, (unsigned long)size>>1); | |
738 | else | |
739 | size = u_size; | |
740 | } | |
741 | ||
742 | return size; | |
743 | } | |
744 | ||
745 | /** | |
746 | * drbd_check_al_size() - Ensures that the AL is of the right size | |
747 | * @mdev: DRBD device. | |
748 | * | |
749 | * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation | |
750 | * failed, and 0 on success. You should call drbd_md_sync() after you called | |
751 | * this function. | |
752 | */ | |
753 | static int drbd_check_al_size(struct drbd_conf *mdev) | |
754 | { | |
755 | struct lru_cache *n, *t; | |
756 | struct lc_element *e; | |
757 | unsigned int in_use; | |
758 | int i; | |
759 | ||
760 | ERR_IF(mdev->sync_conf.al_extents < 7) | |
761 | mdev->sync_conf.al_extents = 127; | |
762 | ||
763 | if (mdev->act_log && | |
764 | mdev->act_log->nr_elements == mdev->sync_conf.al_extents) | |
765 | return 0; | |
766 | ||
767 | in_use = 0; | |
768 | t = mdev->act_log; | |
769 | n = lc_create("act_log", drbd_al_ext_cache, | |
770 | mdev->sync_conf.al_extents, sizeof(struct lc_element), 0); | |
771 | ||
772 | if (n == NULL) { | |
773 | dev_err(DEV, "Cannot allocate act_log lru!\n"); | |
774 | return -ENOMEM; | |
775 | } | |
776 | spin_lock_irq(&mdev->al_lock); | |
777 | if (t) { | |
778 | for (i = 0; i < t->nr_elements; i++) { | |
779 | e = lc_element_by_index(t, i); | |
780 | if (e->refcnt) | |
781 | dev_err(DEV, "refcnt(%d)==%d\n", | |
782 | e->lc_number, e->refcnt); | |
783 | in_use += e->refcnt; | |
784 | } | |
785 | } | |
786 | if (!in_use) | |
787 | mdev->act_log = n; | |
788 | spin_unlock_irq(&mdev->al_lock); | |
789 | if (in_use) { | |
790 | dev_err(DEV, "Activity log still in use!\n"); | |
791 | lc_destroy(n); | |
792 | return -EBUSY; | |
793 | } else { | |
794 | if (t) | |
795 | lc_destroy(t); | |
796 | } | |
797 | drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */ | |
798 | return 0; | |
799 | } | |
800 | ||
99432fcc | 801 | static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size) |
b411b363 PR |
802 | { |
803 | struct request_queue * const q = mdev->rq_queue; | |
99432fcc PR |
804 | int max_hw_sectors = max_bio_size >> 9; |
805 | int max_segments = 0; | |
806 | ||
807 | if (get_ldev_if_state(mdev, D_ATTACHING)) { | |
808 | struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue; | |
809 | ||
810 | max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9); | |
811 | max_segments = mdev->ldev->dc.max_bio_bvecs; | |
812 | put_ldev(mdev); | |
813 | } | |
b411b363 | 814 | |
b411b363 | 815 | blk_queue_logical_block_size(q, 512); |
1816a2b4 LE |
816 | blk_queue_max_hw_sectors(q, max_hw_sectors); |
817 | /* This is the workaround for "bio would need to, but cannot, be split" */ | |
818 | blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS); | |
819 | blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1); | |
b411b363 | 820 | |
99432fcc PR |
821 | if (get_ldev_if_state(mdev, D_ATTACHING)) { |
822 | struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue; | |
823 | ||
824 | blk_queue_stack_limits(q, b); | |
825 | ||
826 | if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) { | |
827 | dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n", | |
828 | q->backing_dev_info.ra_pages, | |
829 | b->backing_dev_info.ra_pages); | |
830 | q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages; | |
831 | } | |
832 | put_ldev(mdev); | |
833 | } | |
834 | } | |
835 | ||
836 | void drbd_reconsider_max_bio_size(struct drbd_conf *mdev) | |
837 | { | |
838 | int now, new, local, peer; | |
839 | ||
840 | now = queue_max_hw_sectors(mdev->rq_queue) << 9; | |
841 | local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */ | |
842 | peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */ | |
b411b363 | 843 | |
99432fcc PR |
844 | if (get_ldev_if_state(mdev, D_ATTACHING)) { |
845 | local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9; | |
846 | mdev->local_max_bio_size = local; | |
847 | put_ldev(mdev); | |
b411b363 | 848 | } |
99432fcc PR |
849 | |
850 | /* We may ignore peer limits if the peer is modern enough. | |
851 | Because new from 8.3.8 onwards the peer can use multiple | |
852 | BIOs for a single peer_request */ | |
853 | if (mdev->state.conn >= C_CONNECTED) { | |
6809384c PR |
854 | if (mdev->agreed_pro_version < 94) { |
855 | peer = min_t(int, mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET); | |
856 | /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */ | |
857 | } else if (mdev->agreed_pro_version == 94) | |
99432fcc PR |
858 | peer = DRBD_MAX_SIZE_H80_PACKET; |
859 | else /* drbd 8.3.8 onwards */ | |
860 | peer = DRBD_MAX_BIO_SIZE; | |
861 | } | |
862 | ||
863 | new = min_t(int, local, peer); | |
864 | ||
865 | if (mdev->state.role == R_PRIMARY && new < now) | |
866 | dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now); | |
867 | ||
868 | if (new != now) | |
869 | dev_info(DEV, "max BIO size = %u\n", new); | |
870 | ||
871 | drbd_setup_queue_param(mdev, new); | |
b411b363 PR |
872 | } |
873 | ||
874 | /* serialize deconfig (worker exiting, doing cleanup) | |
875 | * and reconfig (drbdsetup disk, drbdsetup net) | |
876 | * | |
c518d04f LE |
877 | * Wait for a potentially exiting worker, then restart it, |
878 | * or start a new one. Flush any pending work, there may still be an | |
879 | * after_state_change queued. | |
b411b363 PR |
880 | */ |
881 | static void drbd_reconfig_start(struct drbd_conf *mdev) | |
882 | { | |
6c6c7951 | 883 | wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags)); |
b411b363 PR |
884 | wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags)); |
885 | drbd_thread_start(&mdev->worker); | |
c518d04f | 886 | drbd_flush_workqueue(mdev); |
b411b363 PR |
887 | } |
888 | ||
889 | /* if still unconfigured, stops worker again. | |
890 | * if configured now, clears CONFIG_PENDING. | |
891 | * wakes potential waiters */ | |
892 | static void drbd_reconfig_done(struct drbd_conf *mdev) | |
893 | { | |
894 | spin_lock_irq(&mdev->req_lock); | |
895 | if (mdev->state.disk == D_DISKLESS && | |
896 | mdev->state.conn == C_STANDALONE && | |
897 | mdev->state.role == R_SECONDARY) { | |
898 | set_bit(DEVICE_DYING, &mdev->flags); | |
899 | drbd_thread_stop_nowait(&mdev->worker); | |
900 | } else | |
901 | clear_bit(CONFIG_PENDING, &mdev->flags); | |
902 | spin_unlock_irq(&mdev->req_lock); | |
903 | wake_up(&mdev->state_wait); | |
904 | } | |
905 | ||
0778286a PR |
906 | /* Make sure IO is suspended before calling this function(). */ |
907 | static void drbd_suspend_al(struct drbd_conf *mdev) | |
908 | { | |
909 | int s = 0; | |
910 | ||
911 | if (lc_try_lock(mdev->act_log)) { | |
912 | drbd_al_shrink(mdev); | |
913 | lc_unlock(mdev->act_log); | |
914 | } else { | |
915 | dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n"); | |
916 | return; | |
917 | } | |
918 | ||
919 | spin_lock_irq(&mdev->req_lock); | |
920 | if (mdev->state.conn < C_CONNECTED) | |
921 | s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags); | |
922 | ||
923 | spin_unlock_irq(&mdev->req_lock); | |
924 | ||
925 | if (s) | |
926 | dev_info(DEV, "Suspended AL updates\n"); | |
927 | } | |
928 | ||
b411b363 PR |
929 | /* does always return 0; |
930 | * interesting return code is in reply->ret_code */ | |
931 | static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
932 | struct drbd_nl_cfg_reply *reply) | |
933 | { | |
116676ca | 934 | enum drbd_ret_code retcode; |
b411b363 PR |
935 | enum determine_dev_size dd; |
936 | sector_t max_possible_sectors; | |
937 | sector_t min_md_device_sectors; | |
938 | struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */ | |
e525fd89 | 939 | struct block_device *bdev; |
b411b363 PR |
940 | struct lru_cache *resync_lru = NULL; |
941 | union drbd_state ns, os; | |
f2024e7c | 942 | enum drbd_state_rv rv; |
b411b363 PR |
943 | int cp_discovered = 0; |
944 | int logical_block_size; | |
945 | ||
946 | drbd_reconfig_start(mdev); | |
947 | ||
948 | /* if you want to reconfigure, please tear down first */ | |
949 | if (mdev->state.disk > D_DISKLESS) { | |
950 | retcode = ERR_DISK_CONFIGURED; | |
951 | goto fail; | |
952 | } | |
82f59cc6 LE |
953 | /* It may just now have detached because of IO error. Make sure |
954 | * drbd_ldev_destroy is done already, we may end up here very fast, | |
955 | * e.g. if someone calls attach from the on-io-error handler, | |
956 | * to realize a "hot spare" feature (not that I'd recommend that) */ | |
957 | wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt)); | |
b411b363 | 958 | |
383606e0 LE |
959 | /* make sure there is no leftover from previous force-detach attempts */ |
960 | clear_bit(FORCE_DETACH, &mdev->flags); | |
961 | ||
b411b363 PR |
962 | /* allocation not in the IO path, cqueue thread context */ |
963 | nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL); | |
964 | if (!nbc) { | |
965 | retcode = ERR_NOMEM; | |
966 | goto fail; | |
967 | } | |
968 | ||
969 | nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF; | |
970 | nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF; | |
971 | nbc->dc.fencing = DRBD_FENCING_DEF; | |
972 | nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF; | |
973 | ||
974 | if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) { | |
975 | retcode = ERR_MANDATORY_TAG; | |
976 | goto fail; | |
977 | } | |
978 | ||
979 | if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) { | |
980 | retcode = ERR_MD_IDX_INVALID; | |
981 | goto fail; | |
982 | } | |
983 | ||
47ff2d0a PR |
984 | if (get_net_conf(mdev)) { |
985 | int prot = mdev->net_conf->wire_protocol; | |
986 | put_net_conf(mdev); | |
987 | if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) { | |
988 | retcode = ERR_STONITH_AND_PROT_A; | |
989 | goto fail; | |
990 | } | |
991 | } | |
992 | ||
d4d77629 TH |
993 | bdev = blkdev_get_by_path(nbc->dc.backing_dev, |
994 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev); | |
e525fd89 | 995 | if (IS_ERR(bdev)) { |
b411b363 | 996 | dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev, |
e525fd89 | 997 | PTR_ERR(bdev)); |
b411b363 PR |
998 | retcode = ERR_OPEN_DISK; |
999 | goto fail; | |
1000 | } | |
e525fd89 TH |
1001 | nbc->backing_bdev = bdev; |
1002 | ||
1003 | /* | |
1004 | * meta_dev_idx >= 0: external fixed size, possibly multiple | |
1005 | * drbd sharing one meta device. TODO in that case, paranoia | |
1006 | * check that [md_bdev, meta_dev_idx] is not yet used by some | |
1007 | * other drbd minor! (if you use drbd.conf + drbdadm, that | |
1008 | * should check it for you already; but if you don't, or | |
1009 | * someone fooled it, we need to double check here) | |
1010 | */ | |
d4d77629 TH |
1011 | bdev = blkdev_get_by_path(nbc->dc.meta_dev, |
1012 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, | |
1013 | (nbc->dc.meta_dev_idx < 0) ? | |
1014 | (void *)mdev : (void *)drbd_m_holder); | |
e525fd89 | 1015 | if (IS_ERR(bdev)) { |
b411b363 | 1016 | dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev, |
e525fd89 | 1017 | PTR_ERR(bdev)); |
b411b363 PR |
1018 | retcode = ERR_OPEN_MD_DISK; |
1019 | goto fail; | |
1020 | } | |
e525fd89 | 1021 | nbc->md_bdev = bdev; |
b411b363 | 1022 | |
e525fd89 TH |
1023 | if ((nbc->backing_bdev == nbc->md_bdev) != |
1024 | (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL || | |
1025 | nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) { | |
1026 | retcode = ERR_MD_IDX_INVALID; | |
b411b363 PR |
1027 | goto fail; |
1028 | } | |
1029 | ||
1030 | resync_lru = lc_create("resync", drbd_bm_ext_cache, | |
1031 | 61, sizeof(struct bm_extent), | |
1032 | offsetof(struct bm_extent, lce)); | |
1033 | if (!resync_lru) { | |
1034 | retcode = ERR_NOMEM; | |
e525fd89 | 1035 | goto fail; |
b411b363 PR |
1036 | } |
1037 | ||
1038 | /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */ | |
1039 | drbd_md_set_sector_offsets(mdev, nbc); | |
1040 | ||
1041 | if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) { | |
1042 | dev_err(DEV, "max capacity %llu smaller than disk size %llu\n", | |
1043 | (unsigned long long) drbd_get_max_capacity(nbc), | |
1044 | (unsigned long long) nbc->dc.disk_size); | |
7948bcdc | 1045 | retcode = ERR_DISK_TOO_SMALL; |
e525fd89 | 1046 | goto fail; |
b411b363 PR |
1047 | } |
1048 | ||
1049 | if (nbc->dc.meta_dev_idx < 0) { | |
1050 | max_possible_sectors = DRBD_MAX_SECTORS_FLEX; | |
1051 | /* at least one MB, otherwise it does not make sense */ | |
1052 | min_md_device_sectors = (2<<10); | |
1053 | } else { | |
1054 | max_possible_sectors = DRBD_MAX_SECTORS; | |
1055 | min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1); | |
1056 | } | |
1057 | ||
b411b363 | 1058 | if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) { |
7948bcdc | 1059 | retcode = ERR_MD_DISK_TOO_SMALL; |
b411b363 PR |
1060 | dev_warn(DEV, "refusing attach: md-device too small, " |
1061 | "at least %llu sectors needed for this meta-disk type\n", | |
1062 | (unsigned long long) min_md_device_sectors); | |
e525fd89 | 1063 | goto fail; |
b411b363 PR |
1064 | } |
1065 | ||
1066 | /* Make sure the new disk is big enough | |
1067 | * (we may currently be R_PRIMARY with no local disk...) */ | |
1068 | if (drbd_get_max_capacity(nbc) < | |
1069 | drbd_get_capacity(mdev->this_bdev)) { | |
7948bcdc | 1070 | retcode = ERR_DISK_TOO_SMALL; |
e525fd89 | 1071 | goto fail; |
b411b363 PR |
1072 | } |
1073 | ||
1074 | nbc->known_size = drbd_get_capacity(nbc->backing_bdev); | |
1075 | ||
1352994b LE |
1076 | if (nbc->known_size > max_possible_sectors) { |
1077 | dev_warn(DEV, "==> truncating very big lower level device " | |
1078 | "to currently maximum possible %llu sectors <==\n", | |
1079 | (unsigned long long) max_possible_sectors); | |
1080 | if (nbc->dc.meta_dev_idx >= 0) | |
1081 | dev_warn(DEV, "==>> using internal or flexible " | |
1082 | "meta data may help <<==\n"); | |
1083 | } | |
1084 | ||
b411b363 PR |
1085 | drbd_suspend_io(mdev); |
1086 | /* also wait for the last barrier ack. */ | |
fb22c402 | 1087 | wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state)); |
b411b363 PR |
1088 | /* and for any other previously queued work */ |
1089 | drbd_flush_workqueue(mdev); | |
1090 | ||
f2024e7c AG |
1091 | rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE); |
1092 | retcode = rv; /* FIXME: Type mismatch. */ | |
b411b363 | 1093 | drbd_resume_io(mdev); |
f2024e7c | 1094 | if (rv < SS_SUCCESS) |
e525fd89 | 1095 | goto fail; |
b411b363 PR |
1096 | |
1097 | if (!get_ldev_if_state(mdev, D_ATTACHING)) | |
1098 | goto force_diskless; | |
1099 | ||
1100 | drbd_md_set_sector_offsets(mdev, nbc); | |
1101 | ||
4aa83b7b LE |
1102 | /* allocate a second IO page if logical_block_size != 512 */ |
1103 | logical_block_size = bdev_logical_block_size(nbc->md_bdev); | |
1104 | if (logical_block_size == 0) | |
1105 | logical_block_size = MD_SECTOR_SIZE; | |
1106 | ||
1107 | if (logical_block_size != MD_SECTOR_SIZE) { | |
1108 | if (!mdev->md_io_tmpp) { | |
1109 | struct page *page = alloc_page(GFP_NOIO); | |
1110 | if (!page) | |
1111 | goto force_diskless_dec; | |
1112 | ||
1113 | dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n", | |
1114 | logical_block_size, MD_SECTOR_SIZE); | |
1115 | dev_warn(DEV, "Workaround engaged (has performance impact).\n"); | |
1116 | ||
1117 | mdev->md_io_tmpp = page; | |
1118 | } | |
1119 | } | |
1120 | ||
b411b363 PR |
1121 | if (!mdev->bitmap) { |
1122 | if (drbd_bm_init(mdev)) { | |
1123 | retcode = ERR_NOMEM; | |
1124 | goto force_diskless_dec; | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | retcode = drbd_md_read(mdev, nbc); | |
1129 | if (retcode != NO_ERROR) | |
1130 | goto force_diskless_dec; | |
1131 | ||
1132 | if (mdev->state.conn < C_CONNECTED && | |
1133 | mdev->state.role == R_PRIMARY && | |
1134 | (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) { | |
1135 | dev_err(DEV, "Can only attach to data with current UUID=%016llX\n", | |
1136 | (unsigned long long)mdev->ed_uuid); | |
1137 | retcode = ERR_DATA_NOT_CURRENT; | |
1138 | goto force_diskless_dec; | |
1139 | } | |
1140 | ||
1141 | /* Since we are diskless, fix the activity log first... */ | |
1142 | if (drbd_check_al_size(mdev)) { | |
1143 | retcode = ERR_NOMEM; | |
1144 | goto force_diskless_dec; | |
1145 | } | |
1146 | ||
1147 | /* Prevent shrinking of consistent devices ! */ | |
1148 | if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && | |
a393db6f | 1149 | drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) { |
b411b363 | 1150 | dev_warn(DEV, "refusing to truncate a consistent device\n"); |
7948bcdc | 1151 | retcode = ERR_DISK_TOO_SMALL; |
b411b363 PR |
1152 | goto force_diskless_dec; |
1153 | } | |
1154 | ||
1155 | if (!drbd_al_read_log(mdev, nbc)) { | |
1156 | retcode = ERR_IO_MD_DISK; | |
1157 | goto force_diskless_dec; | |
1158 | } | |
1159 | ||
b411b363 PR |
1160 | /* Reset the "barriers don't work" bits here, then force meta data to |
1161 | * be written, to ensure we determine if barriers are supported. */ | |
1162 | if (nbc->dc.no_md_flush) | |
a8a4e51e | 1163 | set_bit(MD_NO_FUA, &mdev->flags); |
b411b363 | 1164 | else |
a8a4e51e | 1165 | clear_bit(MD_NO_FUA, &mdev->flags); |
b411b363 PR |
1166 | |
1167 | /* Point of no return reached. | |
1168 | * Devices and memory are no longer released by error cleanup below. | |
1169 | * now mdev takes over responsibility, and the state engine should | |
1170 | * clean it up somewhere. */ | |
1171 | D_ASSERT(mdev->ldev == NULL); | |
1172 | mdev->ldev = nbc; | |
1173 | mdev->resync = resync_lru; | |
1174 | nbc = NULL; | |
1175 | resync_lru = NULL; | |
1176 | ||
2451fc3b PR |
1177 | mdev->write_ordering = WO_bdev_flush; |
1178 | drbd_bump_write_ordering(mdev, WO_bdev_flush); | |
b411b363 PR |
1179 | |
1180 | if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY)) | |
1181 | set_bit(CRASHED_PRIMARY, &mdev->flags); | |
1182 | else | |
1183 | clear_bit(CRASHED_PRIMARY, &mdev->flags); | |
1184 | ||
894c6a94 | 1185 | if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && |
fb22c402 | 1186 | !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) { |
b411b363 PR |
1187 | set_bit(CRASHED_PRIMARY, &mdev->flags); |
1188 | cp_discovered = 1; | |
1189 | } | |
1190 | ||
1191 | mdev->send_cnt = 0; | |
1192 | mdev->recv_cnt = 0; | |
1193 | mdev->read_cnt = 0; | |
1194 | mdev->writ_cnt = 0; | |
1195 | ||
99432fcc | 1196 | drbd_reconsider_max_bio_size(mdev); |
b411b363 PR |
1197 | |
1198 | /* If I am currently not R_PRIMARY, | |
1199 | * but meta data primary indicator is set, | |
1200 | * I just now recover from a hard crash, | |
1201 | * and have been R_PRIMARY before that crash. | |
1202 | * | |
1203 | * Now, if I had no connection before that crash | |
1204 | * (have been degraded R_PRIMARY), chances are that | |
1205 | * I won't find my peer now either. | |
1206 | * | |
1207 | * In that case, and _only_ in that case, | |
1208 | * we use the degr-wfc-timeout instead of the default, | |
1209 | * so we can automatically recover from a crash of a | |
1210 | * degraded but active "cluster" after a certain timeout. | |
1211 | */ | |
1212 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
1213 | if (mdev->state.role != R_PRIMARY && | |
1214 | drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && | |
1215 | !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND)) | |
1216 | set_bit(USE_DEGR_WFC_T, &mdev->flags); | |
1217 | ||
24c4830c | 1218 | dd = drbd_determine_dev_size(mdev, 0); |
b411b363 PR |
1219 | if (dd == dev_size_error) { |
1220 | retcode = ERR_NOMEM_BITMAP; | |
1221 | goto force_diskless_dec; | |
1222 | } else if (dd == grew) | |
1223 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); | |
1224 | ||
1225 | if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) { | |
1226 | dev_info(DEV, "Assuming that all blocks are out of sync " | |
1227 | "(aka FullSync)\n"); | |
20ceb2b2 LE |
1228 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, |
1229 | "set_n_write from attaching", BM_LOCKED_MASK)) { | |
b411b363 PR |
1230 | retcode = ERR_IO_MD_DISK; |
1231 | goto force_diskless_dec; | |
1232 | } | |
1233 | } else { | |
20ceb2b2 LE |
1234 | if (drbd_bitmap_io(mdev, &drbd_bm_read, |
1235 | "read from attaching", BM_LOCKED_MASK) < 0) { | |
b411b363 PR |
1236 | retcode = ERR_IO_MD_DISK; |
1237 | goto force_diskless_dec; | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | if (cp_discovered) { | |
1242 | drbd_al_apply_to_bm(mdev); | |
20ceb2b2 LE |
1243 | if (drbd_bitmap_io(mdev, &drbd_bm_write, |
1244 | "crashed primary apply AL", BM_LOCKED_MASK)) { | |
19f843aa LE |
1245 | retcode = ERR_IO_MD_DISK; |
1246 | goto force_diskless_dec; | |
1247 | } | |
b411b363 PR |
1248 | } |
1249 | ||
0778286a PR |
1250 | if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev)) |
1251 | drbd_suspend_al(mdev); /* IO is still suspended here... */ | |
1252 | ||
b411b363 PR |
1253 | spin_lock_irq(&mdev->req_lock); |
1254 | os = mdev->state; | |
1255 | ns.i = os.i; | |
1256 | /* If MDF_CONSISTENT is not set go into inconsistent state, | |
1257 | otherwise investigate MDF_WasUpToDate... | |
1258 | If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state, | |
1259 | otherwise into D_CONSISTENT state. | |
1260 | */ | |
1261 | if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) { | |
1262 | if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE)) | |
1263 | ns.disk = D_CONSISTENT; | |
1264 | else | |
1265 | ns.disk = D_OUTDATED; | |
1266 | } else { | |
1267 | ns.disk = D_INCONSISTENT; | |
1268 | } | |
1269 | ||
1270 | if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED)) | |
1271 | ns.pdsk = D_OUTDATED; | |
1272 | ||
1273 | if ( ns.disk == D_CONSISTENT && | |
1274 | (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE)) | |
1275 | ns.disk = D_UP_TO_DATE; | |
1276 | ||
1277 | /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND, | |
1278 | MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before | |
1279 | this point, because drbd_request_state() modifies these | |
1280 | flags. */ | |
1281 | ||
1282 | /* In case we are C_CONNECTED postpone any decision on the new disk | |
1283 | state after the negotiation phase. */ | |
1284 | if (mdev->state.conn == C_CONNECTED) { | |
1285 | mdev->new_state_tmp.i = ns.i; | |
1286 | ns.i = os.i; | |
1287 | ns.disk = D_NEGOTIATING; | |
dc66c74d PR |
1288 | |
1289 | /* We expect to receive up-to-date UUIDs soon. | |
1290 | To avoid a race in receive_state, free p_uuid while | |
1291 | holding req_lock. I.e. atomic with the state change */ | |
1292 | kfree(mdev->p_uuid); | |
1293 | mdev->p_uuid = NULL; | |
b411b363 PR |
1294 | } |
1295 | ||
1296 | rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
1297 | ns = mdev->state; | |
1298 | spin_unlock_irq(&mdev->req_lock); | |
1299 | ||
1300 | if (rv < SS_SUCCESS) | |
1301 | goto force_diskless_dec; | |
1302 | ||
1303 | if (mdev->state.role == R_PRIMARY) | |
1304 | mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1; | |
1305 | else | |
1306 | mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; | |
1307 | ||
1308 | drbd_md_mark_dirty(mdev); | |
1309 | drbd_md_sync(mdev); | |
1310 | ||
1311 | kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); | |
1312 | put_ldev(mdev); | |
1313 | reply->ret_code = retcode; | |
1314 | drbd_reconfig_done(mdev); | |
1315 | return 0; | |
1316 | ||
1317 | force_diskless_dec: | |
1318 | put_ldev(mdev); | |
1319 | force_diskless: | |
82f59cc6 | 1320 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 | 1321 | drbd_md_sync(mdev); |
b411b363 PR |
1322 | fail: |
1323 | if (nbc) { | |
e525fd89 TH |
1324 | if (nbc->backing_bdev) |
1325 | blkdev_put(nbc->backing_bdev, | |
1326 | FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
1327 | if (nbc->md_bdev) | |
1328 | blkdev_put(nbc->md_bdev, | |
1329 | FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 PR |
1330 | kfree(nbc); |
1331 | } | |
1332 | lc_destroy(resync_lru); | |
1333 | ||
1334 | reply->ret_code = retcode; | |
1335 | drbd_reconfig_done(mdev); | |
1336 | return 0; | |
1337 | } | |
1338 | ||
82f59cc6 LE |
1339 | /* Detaching the disk is a process in multiple stages. First we need to lock |
1340 | * out application IO, in-flight IO, IO stuck in drbd_al_begin_io. | |
1341 | * Then we transition to D_DISKLESS, and wait for put_ldev() to return all | |
1342 | * internal references as well. | |
1343 | * Only then we have finally detached. */ | |
b411b363 PR |
1344 | static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, |
1345 | struct drbd_nl_cfg_reply *reply) | |
1346 | { | |
9a0d9d03 LE |
1347 | enum drbd_ret_code retcode; |
1348 | int ret; | |
02ee8f95 PR |
1349 | struct detach dt = {}; |
1350 | ||
1351 | if (!detach_from_tags(mdev, nlp->tag_list, &dt)) { | |
1352 | reply->ret_code = ERR_MANDATORY_TAG; | |
1353 | goto out; | |
1354 | } | |
1355 | ||
1356 | if (dt.detach_force) { | |
383606e0 | 1357 | set_bit(FORCE_DETACH, &mdev->flags); |
02ee8f95 PR |
1358 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
1359 | reply->ret_code = SS_SUCCESS; | |
1360 | goto out; | |
1361 | } | |
1362 | ||
82f59cc6 | 1363 | drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */ |
a2e91381 | 1364 | drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */ |
9a0d9d03 | 1365 | retcode = drbd_request_state(mdev, NS(disk, D_FAILED)); |
a2e91381 | 1366 | drbd_md_put_buffer(mdev); |
9a0d9d03 LE |
1367 | /* D_FAILED will transition to DISKLESS. */ |
1368 | ret = wait_event_interruptible(mdev->misc_wait, | |
1369 | mdev->state.disk != D_FAILED); | |
82f59cc6 | 1370 | drbd_resume_io(mdev); |
02ee8f95 | 1371 | |
9b2f61ae | 1372 | if ((int)retcode == (int)SS_IS_DISKLESS) |
9a0d9d03 LE |
1373 | retcode = SS_NOTHING_TO_DO; |
1374 | if (ret) | |
1375 | retcode = ERR_INTR; | |
1376 | reply->ret_code = retcode; | |
02ee8f95 | 1377 | out: |
b411b363 PR |
1378 | return 0; |
1379 | } | |
1380 | ||
1381 | static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
1382 | struct drbd_nl_cfg_reply *reply) | |
1383 | { | |
1384 | int i, ns; | |
116676ca | 1385 | enum drbd_ret_code retcode; |
b411b363 PR |
1386 | struct net_conf *new_conf = NULL; |
1387 | struct crypto_hash *tfm = NULL; | |
1388 | struct crypto_hash *integrity_w_tfm = NULL; | |
1389 | struct crypto_hash *integrity_r_tfm = NULL; | |
1390 | struct hlist_head *new_tl_hash = NULL; | |
1391 | struct hlist_head *new_ee_hash = NULL; | |
1392 | struct drbd_conf *odev; | |
1393 | char hmac_name[CRYPTO_MAX_ALG_NAME]; | |
1394 | void *int_dig_out = NULL; | |
1395 | void *int_dig_in = NULL; | |
1396 | void *int_dig_vv = NULL; | |
1397 | struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr; | |
1398 | ||
1399 | drbd_reconfig_start(mdev); | |
1400 | ||
1401 | if (mdev->state.conn > C_STANDALONE) { | |
1402 | retcode = ERR_NET_CONFIGURED; | |
1403 | goto fail; | |
1404 | } | |
1405 | ||
1406 | /* allocation not in the IO path, cqueue thread context */ | |
2db4e42e | 1407 | new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
b411b363 PR |
1408 | if (!new_conf) { |
1409 | retcode = ERR_NOMEM; | |
1410 | goto fail; | |
1411 | } | |
1412 | ||
b411b363 PR |
1413 | new_conf->timeout = DRBD_TIMEOUT_DEF; |
1414 | new_conf->try_connect_int = DRBD_CONNECT_INT_DEF; | |
1415 | new_conf->ping_int = DRBD_PING_INT_DEF; | |
1416 | new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF; | |
1417 | new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF; | |
1418 | new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF; | |
1419 | new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF; | |
1420 | new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF; | |
1421 | new_conf->ko_count = DRBD_KO_COUNT_DEF; | |
1422 | new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF; | |
1423 | new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF; | |
1424 | new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF; | |
1425 | new_conf->want_lose = 0; | |
1426 | new_conf->two_primaries = 0; | |
1427 | new_conf->wire_protocol = DRBD_PROT_C; | |
1428 | new_conf->ping_timeo = DRBD_PING_TIMEO_DEF; | |
1429 | new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF; | |
422028b1 PR |
1430 | new_conf->on_congestion = DRBD_ON_CONGESTION_DEF; |
1431 | new_conf->cong_extents = DRBD_CONG_EXTENTS_DEF; | |
b411b363 PR |
1432 | |
1433 | if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) { | |
1434 | retcode = ERR_MANDATORY_TAG; | |
1435 | goto fail; | |
1436 | } | |
1437 | ||
1438 | if (new_conf->two_primaries | |
1439 | && (new_conf->wire_protocol != DRBD_PROT_C)) { | |
1440 | retcode = ERR_NOT_PROTO_C; | |
1441 | goto fail; | |
47ff2d0a PR |
1442 | } |
1443 | ||
1444 | if (get_ldev(mdev)) { | |
1445 | enum drbd_fencing_p fp = mdev->ldev->dc.fencing; | |
1446 | put_ldev(mdev); | |
1447 | if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) { | |
1448 | retcode = ERR_STONITH_AND_PROT_A; | |
1449 | goto fail; | |
1450 | } | |
1451 | } | |
b411b363 | 1452 | |
422028b1 PR |
1453 | if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) { |
1454 | retcode = ERR_CONG_NOT_PROTO_A; | |
1455 | goto fail; | |
1456 | } | |
1457 | ||
b411b363 PR |
1458 | if (mdev->state.role == R_PRIMARY && new_conf->want_lose) { |
1459 | retcode = ERR_DISCARD; | |
1460 | goto fail; | |
1461 | } | |
1462 | ||
1463 | retcode = NO_ERROR; | |
1464 | ||
1465 | new_my_addr = (struct sockaddr *)&new_conf->my_addr; | |
1466 | new_peer_addr = (struct sockaddr *)&new_conf->peer_addr; | |
1467 | for (i = 0; i < minor_count; i++) { | |
1468 | odev = minor_to_mdev(i); | |
1469 | if (!odev || odev == mdev) | |
1470 | continue; | |
1471 | if (get_net_conf(odev)) { | |
1472 | taken_addr = (struct sockaddr *)&odev->net_conf->my_addr; | |
1473 | if (new_conf->my_addr_len == odev->net_conf->my_addr_len && | |
1474 | !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len)) | |
1475 | retcode = ERR_LOCAL_ADDR; | |
1476 | ||
1477 | taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr; | |
1478 | if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len && | |
1479 | !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len)) | |
1480 | retcode = ERR_PEER_ADDR; | |
1481 | ||
1482 | put_net_conf(odev); | |
1483 | if (retcode != NO_ERROR) | |
1484 | goto fail; | |
1485 | } | |
1486 | } | |
1487 | ||
1488 | if (new_conf->cram_hmac_alg[0] != 0) { | |
1489 | snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", | |
1490 | new_conf->cram_hmac_alg); | |
1491 | tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC); | |
1492 | if (IS_ERR(tfm)) { | |
1493 | tfm = NULL; | |
1494 | retcode = ERR_AUTH_ALG; | |
1495 | goto fail; | |
1496 | } | |
1497 | ||
0798219f | 1498 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { |
b411b363 PR |
1499 | retcode = ERR_AUTH_ALG_ND; |
1500 | goto fail; | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | if (new_conf->integrity_alg[0]) { | |
1505 | integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); | |
1506 | if (IS_ERR(integrity_w_tfm)) { | |
1507 | integrity_w_tfm = NULL; | |
1508 | retcode=ERR_INTEGRITY_ALG; | |
1509 | goto fail; | |
1510 | } | |
1511 | ||
1512 | if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) { | |
1513 | retcode=ERR_INTEGRITY_ALG_ND; | |
1514 | goto fail; | |
1515 | } | |
1516 | ||
1517 | integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); | |
1518 | if (IS_ERR(integrity_r_tfm)) { | |
1519 | integrity_r_tfm = NULL; | |
1520 | retcode=ERR_INTEGRITY_ALG; | |
1521 | goto fail; | |
1522 | } | |
1523 | } | |
1524 | ||
1525 | ns = new_conf->max_epoch_size/8; | |
1526 | if (mdev->tl_hash_s != ns) { | |
1527 | new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); | |
1528 | if (!new_tl_hash) { | |
1529 | retcode = ERR_NOMEM; | |
1530 | goto fail; | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | ns = new_conf->max_buffers/8; | |
1535 | if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) { | |
1536 | new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); | |
1537 | if (!new_ee_hash) { | |
1538 | retcode = ERR_NOMEM; | |
1539 | goto fail; | |
1540 | } | |
1541 | } | |
1542 | ||
1543 | ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0; | |
1544 | ||
1545 | if (integrity_w_tfm) { | |
1546 | i = crypto_hash_digestsize(integrity_w_tfm); | |
1547 | int_dig_out = kmalloc(i, GFP_KERNEL); | |
1548 | if (!int_dig_out) { | |
1549 | retcode = ERR_NOMEM; | |
1550 | goto fail; | |
1551 | } | |
1552 | int_dig_in = kmalloc(i, GFP_KERNEL); | |
1553 | if (!int_dig_in) { | |
1554 | retcode = ERR_NOMEM; | |
1555 | goto fail; | |
1556 | } | |
1557 | int_dig_vv = kmalloc(i, GFP_KERNEL); | |
1558 | if (!int_dig_vv) { | |
1559 | retcode = ERR_NOMEM; | |
1560 | goto fail; | |
1561 | } | |
1562 | } | |
1563 | ||
1564 | if (!mdev->bitmap) { | |
1565 | if(drbd_bm_init(mdev)) { | |
1566 | retcode = ERR_NOMEM; | |
1567 | goto fail; | |
1568 | } | |
1569 | } | |
1570 | ||
f70b3511 | 1571 | drbd_flush_workqueue(mdev); |
b411b363 PR |
1572 | spin_lock_irq(&mdev->req_lock); |
1573 | if (mdev->net_conf != NULL) { | |
1574 | retcode = ERR_NET_CONFIGURED; | |
1575 | spin_unlock_irq(&mdev->req_lock); | |
1576 | goto fail; | |
1577 | } | |
1578 | mdev->net_conf = new_conf; | |
1579 | ||
1580 | mdev->send_cnt = 0; | |
1581 | mdev->recv_cnt = 0; | |
1582 | ||
1583 | if (new_tl_hash) { | |
1584 | kfree(mdev->tl_hash); | |
1585 | mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8; | |
1586 | mdev->tl_hash = new_tl_hash; | |
1587 | } | |
1588 | ||
1589 | if (new_ee_hash) { | |
1590 | kfree(mdev->ee_hash); | |
1591 | mdev->ee_hash_s = mdev->net_conf->max_buffers/8; | |
1592 | mdev->ee_hash = new_ee_hash; | |
1593 | } | |
1594 | ||
1595 | crypto_free_hash(mdev->cram_hmac_tfm); | |
1596 | mdev->cram_hmac_tfm = tfm; | |
1597 | ||
1598 | crypto_free_hash(mdev->integrity_w_tfm); | |
1599 | mdev->integrity_w_tfm = integrity_w_tfm; | |
1600 | ||
1601 | crypto_free_hash(mdev->integrity_r_tfm); | |
1602 | mdev->integrity_r_tfm = integrity_r_tfm; | |
1603 | ||
1604 | kfree(mdev->int_dig_out); | |
1605 | kfree(mdev->int_dig_in); | |
1606 | kfree(mdev->int_dig_vv); | |
1607 | mdev->int_dig_out=int_dig_out; | |
1608 | mdev->int_dig_in=int_dig_in; | |
1609 | mdev->int_dig_vv=int_dig_vv; | |
f70b3511 | 1610 | retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL); |
b411b363 PR |
1611 | spin_unlock_irq(&mdev->req_lock); |
1612 | ||
b411b363 PR |
1613 | kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); |
1614 | reply->ret_code = retcode; | |
1615 | drbd_reconfig_done(mdev); | |
1616 | return 0; | |
1617 | ||
1618 | fail: | |
1619 | kfree(int_dig_out); | |
1620 | kfree(int_dig_in); | |
1621 | kfree(int_dig_vv); | |
1622 | crypto_free_hash(tfm); | |
1623 | crypto_free_hash(integrity_w_tfm); | |
1624 | crypto_free_hash(integrity_r_tfm); | |
1625 | kfree(new_tl_hash); | |
1626 | kfree(new_ee_hash); | |
1627 | kfree(new_conf); | |
1628 | ||
1629 | reply->ret_code = retcode; | |
1630 | drbd_reconfig_done(mdev); | |
1631 | return 0; | |
1632 | } | |
1633 | ||
1634 | static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
1635 | struct drbd_nl_cfg_reply *reply) | |
1636 | { | |
1637 | int retcode; | |
2561b9c1 PR |
1638 | struct disconnect dc; |
1639 | ||
1640 | memset(&dc, 0, sizeof(struct disconnect)); | |
1641 | if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) { | |
1642 | retcode = ERR_MANDATORY_TAG; | |
1643 | goto fail; | |
1644 | } | |
1645 | ||
1646 | if (dc.force) { | |
1647 | spin_lock_irq(&mdev->req_lock); | |
1648 | if (mdev->state.conn >= C_WF_CONNECTION) | |
1649 | _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL); | |
1650 | spin_unlock_irq(&mdev->req_lock); | |
1651 | goto done; | |
1652 | } | |
b411b363 PR |
1653 | |
1654 | retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED); | |
1655 | ||
1656 | if (retcode == SS_NOTHING_TO_DO) | |
1657 | goto done; | |
1658 | else if (retcode == SS_ALREADY_STANDALONE) | |
1659 | goto done; | |
1660 | else if (retcode == SS_PRIMARY_NOP) { | |
1661 | /* Our statche checking code wants to see the peer outdated. */ | |
1662 | retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, | |
1663 | pdsk, D_OUTDATED)); | |
1664 | } else if (retcode == SS_CW_FAILED_BY_PEER) { | |
1665 | /* The peer probably wants to see us outdated. */ | |
1666 | retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, | |
1667 | disk, D_OUTDATED), | |
1668 | CS_ORDERED); | |
1669 | if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) { | |
1670 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
1671 | retcode = SS_SUCCESS; | |
1672 | } | |
1673 | } | |
1674 | ||
1675 | if (retcode < SS_SUCCESS) | |
1676 | goto fail; | |
1677 | ||
1678 | if (wait_event_interruptible(mdev->state_wait, | |
1679 | mdev->state.conn != C_DISCONNECTING)) { | |
1680 | /* Do not test for mdev->state.conn == C_STANDALONE, since | |
1681 | someone else might connect us in the mean time! */ | |
1682 | retcode = ERR_INTR; | |
1683 | goto fail; | |
1684 | } | |
1685 | ||
1686 | done: | |
1687 | retcode = NO_ERROR; | |
1688 | fail: | |
1689 | drbd_md_sync(mdev); | |
1690 | reply->ret_code = retcode; | |
1691 | return 0; | |
1692 | } | |
1693 | ||
1694 | void resync_after_online_grow(struct drbd_conf *mdev) | |
1695 | { | |
1696 | int iass; /* I am sync source */ | |
1697 | ||
1698 | dev_info(DEV, "Resync of new storage after online grow\n"); | |
1699 | if (mdev->state.role != mdev->state.peer) | |
1700 | iass = (mdev->state.role == R_PRIMARY); | |
1701 | else | |
1702 | iass = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
1703 | ||
1704 | if (iass) | |
1705 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1706 | else | |
1707 | _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE); | |
1708 | } | |
1709 | ||
1710 | static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
1711 | struct drbd_nl_cfg_reply *reply) | |
1712 | { | |
1713 | struct resize rs; | |
1714 | int retcode = NO_ERROR; | |
b411b363 | 1715 | enum determine_dev_size dd; |
6495d2c6 | 1716 | enum dds_flags ddsf; |
b411b363 PR |
1717 | |
1718 | memset(&rs, 0, sizeof(struct resize)); | |
1719 | if (!resize_from_tags(mdev, nlp->tag_list, &rs)) { | |
1720 | retcode = ERR_MANDATORY_TAG; | |
1721 | goto fail; | |
1722 | } | |
1723 | ||
1724 | if (mdev->state.conn > C_CONNECTED) { | |
1725 | retcode = ERR_RESIZE_RESYNC; | |
1726 | goto fail; | |
1727 | } | |
1728 | ||
1729 | if (mdev->state.role == R_SECONDARY && | |
1730 | mdev->state.peer == R_SECONDARY) { | |
1731 | retcode = ERR_NO_PRIMARY; | |
1732 | goto fail; | |
1733 | } | |
1734 | ||
1735 | if (!get_ldev(mdev)) { | |
1736 | retcode = ERR_NO_DISK; | |
1737 | goto fail; | |
1738 | } | |
1739 | ||
6495d2c6 PR |
1740 | if (rs.no_resync && mdev->agreed_pro_version < 93) { |
1741 | retcode = ERR_NEED_APV_93; | |
7b4e4d31 | 1742 | goto fail_ldev; |
6495d2c6 PR |
1743 | } |
1744 | ||
087c2492 | 1745 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) |
b411b363 | 1746 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); |
b411b363 PR |
1747 | |
1748 | mdev->ldev->dc.disk_size = (sector_t)rs.resize_size; | |
6495d2c6 | 1749 | ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0); |
24c4830c | 1750 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
1751 | drbd_md_sync(mdev); |
1752 | put_ldev(mdev); | |
1753 | if (dd == dev_size_error) { | |
1754 | retcode = ERR_NOMEM_BITMAP; | |
1755 | goto fail; | |
1756 | } | |
1757 | ||
087c2492 | 1758 | if (mdev->state.conn == C_CONNECTED) { |
b411b363 PR |
1759 | if (dd == grew) |
1760 | set_bit(RESIZE_PENDING, &mdev->flags); | |
1761 | ||
1762 | drbd_send_uuids(mdev); | |
6495d2c6 | 1763 | drbd_send_sizes(mdev, 1, ddsf); |
b411b363 PR |
1764 | } |
1765 | ||
1766 | fail: | |
1767 | reply->ret_code = retcode; | |
1768 | return 0; | |
7b4e4d31 AG |
1769 | |
1770 | fail_ldev: | |
1771 | put_ldev(mdev); | |
1772 | goto fail; | |
b411b363 PR |
1773 | } |
1774 | ||
1775 | static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
1776 | struct drbd_nl_cfg_reply *reply) | |
1777 | { | |
1778 | int retcode = NO_ERROR; | |
1779 | int err; | |
1780 | int ovr; /* online verify running */ | |
1781 | int rsr; /* re-sync running */ | |
1782 | struct crypto_hash *verify_tfm = NULL; | |
1783 | struct crypto_hash *csums_tfm = NULL; | |
1784 | struct syncer_conf sc; | |
1785 | cpumask_var_t new_cpu_mask; | |
778f271d PR |
1786 | int *rs_plan_s = NULL; |
1787 | int fifo_size; | |
b411b363 PR |
1788 | |
1789 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) { | |
1790 | retcode = ERR_NOMEM; | |
1791 | goto fail; | |
1792 | } | |
1793 | ||
1794 | if (nlp->flags & DRBD_NL_SET_DEFAULTS) { | |
1795 | memset(&sc, 0, sizeof(struct syncer_conf)); | |
1796 | sc.rate = DRBD_RATE_DEF; | |
1797 | sc.after = DRBD_AFTER_DEF; | |
1798 | sc.al_extents = DRBD_AL_EXTENTS_DEF; | |
265be2d0 | 1799 | sc.on_no_data = DRBD_ON_NO_DATA_DEF; |
9a31d716 PR |
1800 | sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF; |
1801 | sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF; | |
1802 | sc.c_fill_target = DRBD_C_FILL_TARGET_DEF; | |
0f0601f4 LE |
1803 | sc.c_max_rate = DRBD_C_MAX_RATE_DEF; |
1804 | sc.c_min_rate = DRBD_C_MIN_RATE_DEF; | |
b411b363 PR |
1805 | } else |
1806 | memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf)); | |
1807 | ||
1808 | if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) { | |
1809 | retcode = ERR_MANDATORY_TAG; | |
1810 | goto fail; | |
1811 | } | |
1812 | ||
1813 | /* re-sync running */ | |
1814 | rsr = ( mdev->state.conn == C_SYNC_SOURCE || | |
1815 | mdev->state.conn == C_SYNC_TARGET || | |
1816 | mdev->state.conn == C_PAUSED_SYNC_S || | |
1817 | mdev->state.conn == C_PAUSED_SYNC_T ); | |
1818 | ||
1819 | if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) { | |
1820 | retcode = ERR_CSUMS_RESYNC_RUNNING; | |
1821 | goto fail; | |
1822 | } | |
1823 | ||
1824 | if (!rsr && sc.csums_alg[0]) { | |
1825 | csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC); | |
1826 | if (IS_ERR(csums_tfm)) { | |
1827 | csums_tfm = NULL; | |
1828 | retcode = ERR_CSUMS_ALG; | |
1829 | goto fail; | |
1830 | } | |
1831 | ||
1832 | if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) { | |
1833 | retcode = ERR_CSUMS_ALG_ND; | |
1834 | goto fail; | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | /* online verify running */ | |
1839 | ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T); | |
1840 | ||
1841 | if (ovr) { | |
1842 | if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) { | |
1843 | retcode = ERR_VERIFY_RUNNING; | |
1844 | goto fail; | |
1845 | } | |
1846 | } | |
1847 | ||
1848 | if (!ovr && sc.verify_alg[0]) { | |
1849 | verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC); | |
1850 | if (IS_ERR(verify_tfm)) { | |
1851 | verify_tfm = NULL; | |
1852 | retcode = ERR_VERIFY_ALG; | |
1853 | goto fail; | |
1854 | } | |
1855 | ||
1856 | if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) { | |
1857 | retcode = ERR_VERIFY_ALG_ND; | |
1858 | goto fail; | |
1859 | } | |
1860 | } | |
1861 | ||
1862 | /* silently ignore cpu mask on UP kernel */ | |
1863 | if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) { | |
ddad9ef5 | 1864 | err = bitmap_parse(sc.cpu_mask, 32, |
b411b363 PR |
1865 | cpumask_bits(new_cpu_mask), nr_cpu_ids); |
1866 | if (err) { | |
ddad9ef5 | 1867 | dev_warn(DEV, "bitmap_parse() failed with %d\n", err); |
b411b363 PR |
1868 | retcode = ERR_CPU_MASK_PARSE; |
1869 | goto fail; | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | ERR_IF (sc.rate < 1) sc.rate = 1; | |
1874 | ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */ | |
1875 | #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT) | |
1876 | if (sc.al_extents > AL_MAX) { | |
1877 | dev_err(DEV, "sc.al_extents > %d\n", AL_MAX); | |
1878 | sc.al_extents = AL_MAX; | |
1879 | } | |
1880 | #undef AL_MAX | |
1881 | ||
ef50a3e3 LE |
1882 | /* to avoid spurious errors when configuring minors before configuring |
1883 | * the minors they depend on: if necessary, first create the minor we | |
1884 | * depend on */ | |
1885 | if (sc.after >= 0) | |
1886 | ensure_mdev(sc.after, 1); | |
1887 | ||
b411b363 PR |
1888 | /* most sanity checks done, try to assign the new sync-after |
1889 | * dependency. need to hold the global lock in there, | |
1890 | * to avoid a race in the dependency loop check. */ | |
1891 | retcode = drbd_alter_sa(mdev, sc.after); | |
1892 | if (retcode != NO_ERROR) | |
1893 | goto fail; | |
1894 | ||
778f271d PR |
1895 | fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
1896 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
1897 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
1898 | if (!rs_plan_s) { | |
1899 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
1900 | retcode = ERR_NOMEM; | |
1901 | goto fail; | |
1902 | } | |
1903 | } | |
1904 | ||
b411b363 PR |
1905 | /* ok, assign the rest of it as well. |
1906 | * lock against receive_SyncParam() */ | |
1907 | spin_lock(&mdev->peer_seq_lock); | |
1908 | mdev->sync_conf = sc; | |
1909 | ||
1910 | if (!rsr) { | |
1911 | crypto_free_hash(mdev->csums_tfm); | |
1912 | mdev->csums_tfm = csums_tfm; | |
1913 | csums_tfm = NULL; | |
1914 | } | |
1915 | ||
1916 | if (!ovr) { | |
1917 | crypto_free_hash(mdev->verify_tfm); | |
1918 | mdev->verify_tfm = verify_tfm; | |
1919 | verify_tfm = NULL; | |
1920 | } | |
778f271d PR |
1921 | |
1922 | if (fifo_size != mdev->rs_plan_s.size) { | |
1923 | kfree(mdev->rs_plan_s.values); | |
1924 | mdev->rs_plan_s.values = rs_plan_s; | |
1925 | mdev->rs_plan_s.size = fifo_size; | |
1926 | mdev->rs_planed = 0; | |
1927 | rs_plan_s = NULL; | |
1928 | } | |
1929 | ||
b411b363 PR |
1930 | spin_unlock(&mdev->peer_seq_lock); |
1931 | ||
1932 | if (get_ldev(mdev)) { | |
1933 | wait_event(mdev->al_wait, lc_try_lock(mdev->act_log)); | |
1934 | drbd_al_shrink(mdev); | |
1935 | err = drbd_check_al_size(mdev); | |
1936 | lc_unlock(mdev->act_log); | |
1937 | wake_up(&mdev->al_wait); | |
1938 | ||
1939 | put_ldev(mdev); | |
1940 | drbd_md_sync(mdev); | |
1941 | ||
1942 | if (err) { | |
1943 | retcode = ERR_NOMEM; | |
1944 | goto fail; | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | if (mdev->state.conn >= C_CONNECTED) | |
1949 | drbd_send_sync_param(mdev, &sc); | |
1950 | ||
1951 | if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) { | |
1952 | cpumask_copy(mdev->cpu_mask, new_cpu_mask); | |
1953 | drbd_calc_cpu_mask(mdev); | |
1954 | mdev->receiver.reset_cpu_mask = 1; | |
1955 | mdev->asender.reset_cpu_mask = 1; | |
1956 | mdev->worker.reset_cpu_mask = 1; | |
1957 | } | |
1958 | ||
1959 | kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); | |
1960 | fail: | |
778f271d | 1961 | kfree(rs_plan_s); |
b411b363 PR |
1962 | free_cpumask_var(new_cpu_mask); |
1963 | crypto_free_hash(csums_tfm); | |
1964 | crypto_free_hash(verify_tfm); | |
1965 | reply->ret_code = retcode; | |
1966 | return 0; | |
1967 | } | |
1968 | ||
1969 | static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
1970 | struct drbd_nl_cfg_reply *reply) | |
1971 | { | |
1972 | int retcode; | |
1973 | ||
194bfb32 LE |
1974 | /* If there is still bitmap IO pending, probably because of a previous |
1975 | * resync just being finished, wait for it before requesting a new resync. */ | |
a574daf5 | 1976 | drbd_suspend_io(mdev); |
194bfb32 LE |
1977 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); |
1978 | ||
b411b363 PR |
1979 | retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED); |
1980 | ||
1981 | if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION) | |
1982 | retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); | |
1983 | ||
1984 | while (retcode == SS_NEED_CONNECTION) { | |
1985 | spin_lock_irq(&mdev->req_lock); | |
1986 | if (mdev->state.conn < C_CONNECTED) | |
1987 | retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL); | |
1988 | spin_unlock_irq(&mdev->req_lock); | |
1989 | ||
1990 | if (retcode != SS_NEED_CONNECTION) | |
1991 | break; | |
1992 | ||
1993 | retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); | |
1994 | } | |
a574daf5 | 1995 | drbd_resume_io(mdev); |
b411b363 PR |
1996 | |
1997 | reply->ret_code = retcode; | |
1998 | return 0; | |
1999 | } | |
2000 | ||
0778286a PR |
2001 | static int drbd_bmio_set_susp_al(struct drbd_conf *mdev) |
2002 | { | |
2003 | int rv; | |
2004 | ||
2005 | rv = drbd_bmio_set_n_write(mdev); | |
2006 | drbd_suspend_al(mdev); | |
2007 | return rv; | |
2008 | } | |
2009 | ||
b411b363 PR |
2010 | static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, |
2011 | struct drbd_nl_cfg_reply *reply) | |
2012 | { | |
0778286a | 2013 | int retcode; |
b411b363 | 2014 | |
194bfb32 LE |
2015 | /* If there is still bitmap IO pending, probably because of a previous |
2016 | * resync just being finished, wait for it before requesting a new resync. */ | |
a574daf5 | 2017 | drbd_suspend_io(mdev); |
194bfb32 LE |
2018 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); |
2019 | ||
0778286a PR |
2020 | retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED); |
2021 | ||
2022 | if (retcode < SS_SUCCESS) { | |
2023 | if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) { | |
2024 | /* The peer will get a resync upon connect anyways. Just make that | |
2025 | into a full resync. */ | |
2026 | retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT)); | |
2027 | if (retcode >= SS_SUCCESS) { | |
0778286a | 2028 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al, |
20ceb2b2 LE |
2029 | "set_n_write from invalidate_peer", |
2030 | BM_LOCKED_SET_ALLOWED)) | |
0778286a PR |
2031 | retcode = ERR_IO_MD_DISK; |
2032 | } | |
2033 | } else | |
2034 | retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S)); | |
2035 | } | |
a574daf5 | 2036 | drbd_resume_io(mdev); |
b411b363 | 2037 | |
0778286a | 2038 | reply->ret_code = retcode; |
b411b363 PR |
2039 | return 0; |
2040 | } | |
2041 | ||
2042 | static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2043 | struct drbd_nl_cfg_reply *reply) | |
2044 | { | |
2045 | int retcode = NO_ERROR; | |
2046 | ||
2047 | if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO) | |
2048 | retcode = ERR_PAUSE_IS_SET; | |
2049 | ||
2050 | reply->ret_code = retcode; | |
2051 | return 0; | |
2052 | } | |
2053 | ||
2054 | static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2055 | struct drbd_nl_cfg_reply *reply) | |
2056 | { | |
2057 | int retcode = NO_ERROR; | |
cd88d030 | 2058 | union drbd_state s; |
b411b363 | 2059 | |
cd88d030 PR |
2060 | if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) { |
2061 | s = mdev->state; | |
2062 | if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) { | |
2063 | retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP : | |
2064 | s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR; | |
2065 | } else { | |
2066 | retcode = ERR_PAUSE_IS_CLEAR; | |
2067 | } | |
2068 | } | |
b411b363 PR |
2069 | |
2070 | reply->ret_code = retcode; | |
2071 | return 0; | |
2072 | } | |
2073 | ||
2074 | static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2075 | struct drbd_nl_cfg_reply *reply) | |
2076 | { | |
2077 | reply->ret_code = drbd_request_state(mdev, NS(susp, 1)); | |
2078 | ||
2079 | return 0; | |
2080 | } | |
2081 | ||
2082 | static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2083 | struct drbd_nl_cfg_reply *reply) | |
2084 | { | |
43a5182c PR |
2085 | if (test_bit(NEW_CUR_UUID, &mdev->flags)) { |
2086 | drbd_uuid_new_current(mdev); | |
2087 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
43a5182c | 2088 | } |
265be2d0 | 2089 | drbd_suspend_io(mdev); |
fb22c402 | 2090 | reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0)); |
265be2d0 PR |
2091 | if (reply->ret_code == SS_SUCCESS) { |
2092 | if (mdev->state.conn < C_CONNECTED) | |
2093 | tl_clear(mdev); | |
2094 | if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED) | |
2095 | tl_restart(mdev, fail_frozen_disk_io); | |
2096 | } | |
2097 | drbd_resume_io(mdev); | |
2098 | ||
b411b363 PR |
2099 | return 0; |
2100 | } | |
2101 | ||
2102 | static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2103 | struct drbd_nl_cfg_reply *reply) | |
2104 | { | |
2105 | reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED)); | |
2106 | return 0; | |
2107 | } | |
2108 | ||
2109 | static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2110 | struct drbd_nl_cfg_reply *reply) | |
2111 | { | |
2112 | unsigned short *tl; | |
2113 | ||
2114 | tl = reply->tag_list; | |
2115 | ||
2116 | if (get_ldev(mdev)) { | |
2117 | tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl); | |
2118 | put_ldev(mdev); | |
2119 | } | |
2120 | ||
2121 | if (get_net_conf(mdev)) { | |
2122 | tl = net_conf_to_tags(mdev, mdev->net_conf, tl); | |
2123 | put_net_conf(mdev); | |
2124 | } | |
2125 | tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl); | |
2126 | ||
2127 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2128 | ||
2129 | return (int)((char *)tl - (char *)reply->tag_list); | |
2130 | } | |
2131 | ||
2132 | static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2133 | struct drbd_nl_cfg_reply *reply) | |
2134 | { | |
2135 | unsigned short *tl = reply->tag_list; | |
2136 | union drbd_state s = mdev->state; | |
2137 | unsigned long rs_left; | |
2138 | unsigned int res; | |
2139 | ||
2140 | tl = get_state_to_tags(mdev, (struct get_state *)&s, tl); | |
2141 | ||
2142 | /* no local ref, no bitmap, no syncer progress. */ | |
2143 | if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) { | |
2144 | if (get_ldev(mdev)) { | |
2145 | drbd_get_syncer_progress(mdev, &rs_left, &res); | |
2146 | tl = tl_add_int(tl, T_sync_progress, &res); | |
2147 | put_ldev(mdev); | |
2148 | } | |
2149 | } | |
2150 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2151 | ||
2152 | return (int)((char *)tl - (char *)reply->tag_list); | |
2153 | } | |
2154 | ||
2155 | static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2156 | struct drbd_nl_cfg_reply *reply) | |
2157 | { | |
2158 | unsigned short *tl; | |
2159 | ||
2160 | tl = reply->tag_list; | |
2161 | ||
2162 | if (get_ldev(mdev)) { | |
2163 | tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64)); | |
2164 | tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags); | |
2165 | put_ldev(mdev); | |
2166 | } | |
2167 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2168 | ||
2169 | return (int)((char *)tl - (char *)reply->tag_list); | |
2170 | } | |
2171 | ||
2172 | /** | |
2173 | * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use | |
2174 | * @mdev: DRBD device. | |
2175 | * @nlp: Netlink/connector packet from drbdsetup | |
2176 | * @reply: Reply packet for drbdsetup | |
2177 | */ | |
2178 | static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2179 | struct drbd_nl_cfg_reply *reply) | |
2180 | { | |
2181 | unsigned short *tl; | |
2182 | char rv; | |
2183 | ||
2184 | tl = reply->tag_list; | |
2185 | ||
2186 | rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED : | |
2187 | test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT; | |
2188 | ||
2189 | tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv)); | |
2190 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2191 | ||
2192 | return (int)((char *)tl - (char *)reply->tag_list); | |
2193 | } | |
2194 | ||
2195 | static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2196 | struct drbd_nl_cfg_reply *reply) | |
2197 | { | |
2198 | /* default to resume from last known position, if possible */ | |
2199 | struct start_ov args = | |
2200 | { .start_sector = mdev->ov_start_sector }; | |
2201 | ||
2202 | if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) { | |
2203 | reply->ret_code = ERR_MANDATORY_TAG; | |
2204 | return 0; | |
2205 | } | |
873b0d5f LE |
2206 | |
2207 | /* If there is still bitmap IO pending, e.g. previous resync or verify | |
2208 | * just being finished, wait for it before requesting a new resync. */ | |
a574daf5 | 2209 | drbd_suspend_io(mdev); |
873b0d5f LE |
2210 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); |
2211 | ||
b411b363 PR |
2212 | /* w_make_ov_request expects position to be aligned */ |
2213 | mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT; | |
2214 | reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S)); | |
a574daf5 | 2215 | drbd_resume_io(mdev); |
b411b363 PR |
2216 | return 0; |
2217 | } | |
2218 | ||
2219 | ||
2220 | static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, | |
2221 | struct drbd_nl_cfg_reply *reply) | |
2222 | { | |
2223 | int retcode = NO_ERROR; | |
2224 | int skip_initial_sync = 0; | |
2225 | int err; | |
2226 | ||
2227 | struct new_c_uuid args; | |
2228 | ||
2229 | memset(&args, 0, sizeof(struct new_c_uuid)); | |
2230 | if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) { | |
2231 | reply->ret_code = ERR_MANDATORY_TAG; | |
2232 | return 0; | |
2233 | } | |
2234 | ||
2235 | mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */ | |
2236 | ||
2237 | if (!get_ldev(mdev)) { | |
2238 | retcode = ERR_NO_DISK; | |
2239 | goto out; | |
2240 | } | |
2241 | ||
2242 | /* this is "skip initial sync", assume to be clean */ | |
2243 | if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 && | |
2244 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) { | |
2245 | dev_info(DEV, "Preparing to skip initial sync\n"); | |
2246 | skip_initial_sync = 1; | |
2247 | } else if (mdev->state.conn != C_STANDALONE) { | |
2248 | retcode = ERR_CONNECTED; | |
2249 | goto out_dec; | |
2250 | } | |
2251 | ||
2252 | drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */ | |
2253 | drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */ | |
2254 | ||
2255 | if (args.clear_bm) { | |
20ceb2b2 LE |
2256 | err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, |
2257 | "clear_n_write from new_c_uuid", BM_LOCKED_MASK); | |
b411b363 PR |
2258 | if (err) { |
2259 | dev_err(DEV, "Writing bitmap failed with %d\n",err); | |
2260 | retcode = ERR_IO_MD_DISK; | |
2261 | } | |
2262 | if (skip_initial_sync) { | |
2263 | drbd_send_uuids_skip_initial_sync(mdev); | |
2264 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
62b0da3a | 2265 | drbd_print_uuids(mdev, "cleared bitmap UUID"); |
b411b363 PR |
2266 | spin_lock_irq(&mdev->req_lock); |
2267 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
2268 | CS_VERBOSE, NULL); | |
2269 | spin_unlock_irq(&mdev->req_lock); | |
2270 | } | |
2271 | } | |
2272 | ||
2273 | drbd_md_sync(mdev); | |
2274 | out_dec: | |
2275 | put_ldev(mdev); | |
2276 | out: | |
2277 | mutex_unlock(&mdev->state_mutex); | |
2278 | ||
2279 | reply->ret_code = retcode; | |
2280 | return 0; | |
2281 | } | |
2282 | ||
b411b363 PR |
2283 | struct cn_handler_struct { |
2284 | int (*function)(struct drbd_conf *, | |
2285 | struct drbd_nl_cfg_req *, | |
2286 | struct drbd_nl_cfg_reply *); | |
2287 | int reply_body_size; | |
2288 | }; | |
2289 | ||
2290 | static struct cn_handler_struct cnd_table[] = { | |
2291 | [ P_primary ] = { &drbd_nl_primary, 0 }, | |
2292 | [ P_secondary ] = { &drbd_nl_secondary, 0 }, | |
2293 | [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 }, | |
2294 | [ P_detach ] = { &drbd_nl_detach, 0 }, | |
2295 | [ P_net_conf ] = { &drbd_nl_net_conf, 0 }, | |
2296 | [ P_disconnect ] = { &drbd_nl_disconnect, 0 }, | |
2297 | [ P_resize ] = { &drbd_nl_resize, 0 }, | |
2298 | [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 }, | |
2299 | [ P_invalidate ] = { &drbd_nl_invalidate, 0 }, | |
2300 | [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 }, | |
2301 | [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 }, | |
2302 | [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 }, | |
2303 | [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 }, | |
2304 | [ P_resume_io ] = { &drbd_nl_resume_io, 0 }, | |
2305 | [ P_outdate ] = { &drbd_nl_outdate, 0 }, | |
2306 | [ P_get_config ] = { &drbd_nl_get_config, | |
2307 | sizeof(struct syncer_conf_tag_len_struct) + | |
2308 | sizeof(struct disk_conf_tag_len_struct) + | |
2309 | sizeof(struct net_conf_tag_len_struct) }, | |
2310 | [ P_get_state ] = { &drbd_nl_get_state, | |
2311 | sizeof(struct get_state_tag_len_struct) + | |
2312 | sizeof(struct sync_progress_tag_len_struct) }, | |
2313 | [ P_get_uuids ] = { &drbd_nl_get_uuids, | |
2314 | sizeof(struct get_uuids_tag_len_struct) }, | |
2315 | [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag, | |
2316 | sizeof(struct get_timeout_flag_tag_len_struct)}, | |
2317 | [ P_start_ov ] = { &drbd_nl_start_ov, 0 }, | |
2318 | [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 }, | |
2319 | }; | |
2320 | ||
9f5180e5 | 2321 | static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp) |
b411b363 PR |
2322 | { |
2323 | struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data; | |
2324 | struct cn_handler_struct *cm; | |
2325 | struct cn_msg *cn_reply; | |
2326 | struct drbd_nl_cfg_reply *reply; | |
2327 | struct drbd_conf *mdev; | |
2328 | int retcode, rr; | |
2329 | int reply_size = sizeof(struct cn_msg) | |
2330 | + sizeof(struct drbd_nl_cfg_reply) | |
2331 | + sizeof(short int); | |
2332 | ||
2333 | if (!try_module_get(THIS_MODULE)) { | |
2334 | printk(KERN_ERR "drbd: try_module_get() failed!\n"); | |
2335 | return; | |
2336 | } | |
2337 | ||
38bf1953 | 2338 | if (!capable(CAP_SYS_ADMIN)) { |
9f5180e5 PR |
2339 | retcode = ERR_PERM; |
2340 | goto fail; | |
2341 | } | |
2342 | ||
ef50a3e3 LE |
2343 | mdev = ensure_mdev(nlp->drbd_minor, |
2344 | (nlp->flags & DRBD_NL_CREATE_DEVICE)); | |
b411b363 PR |
2345 | if (!mdev) { |
2346 | retcode = ERR_MINOR_INVALID; | |
2347 | goto fail; | |
2348 | } | |
2349 | ||
42ff269d LE |
2350 | if (nlp->packet_type >= P_nl_after_last_packet || |
2351 | nlp->packet_type == P_return_code_only) { | |
b411b363 PR |
2352 | retcode = ERR_PACKET_NR; |
2353 | goto fail; | |
2354 | } | |
2355 | ||
2356 | cm = cnd_table + nlp->packet_type; | |
2357 | ||
2358 | /* This may happen if packet number is 0: */ | |
2359 | if (cm->function == NULL) { | |
2360 | retcode = ERR_PACKET_NR; | |
2361 | goto fail; | |
2362 | } | |
2363 | ||
2364 | reply_size += cm->reply_body_size; | |
2365 | ||
2366 | /* allocation not in the IO path, cqueue thread context */ | |
3e3a7766 | 2367 | cn_reply = kzalloc(reply_size, GFP_KERNEL); |
b411b363 PR |
2368 | if (!cn_reply) { |
2369 | retcode = ERR_NOMEM; | |
2370 | goto fail; | |
2371 | } | |
2372 | reply = (struct drbd_nl_cfg_reply *) cn_reply->data; | |
2373 | ||
2374 | reply->packet_type = | |
42ff269d | 2375 | cm->reply_body_size ? nlp->packet_type : P_return_code_only; |
b411b363 PR |
2376 | reply->minor = nlp->drbd_minor; |
2377 | reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */ | |
2378 | /* reply->tag_list; might be modified by cm->function. */ | |
2379 | ||
2380 | rr = cm->function(mdev, nlp, reply); | |
2381 | ||
2382 | cn_reply->id = req->id; | |
2383 | cn_reply->seq = req->seq; | |
2384 | cn_reply->ack = req->ack + 1; | |
2385 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr; | |
2386 | cn_reply->flags = 0; | |
2387 | ||
b411b363 PR |
2388 | rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL); |
2389 | if (rr && rr != -ESRCH) | |
2390 | printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); | |
2391 | ||
2392 | kfree(cn_reply); | |
2393 | module_put(THIS_MODULE); | |
2394 | return; | |
2395 | fail: | |
2396 | drbd_nl_send_reply(req, retcode); | |
2397 | module_put(THIS_MODULE); | |
2398 | } | |
2399 | ||
2400 | static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */ | |
2401 | ||
2402 | static unsigned short * | |
2403 | __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, | |
2404 | unsigned short len, int nul_terminated) | |
2405 | { | |
2406 | unsigned short l = tag_descriptions[tag_number(tag)].max_len; | |
2407 | len = (len < l) ? len : l; | |
2408 | put_unaligned(tag, tl++); | |
2409 | put_unaligned(len, tl++); | |
2410 | memcpy(tl, data, len); | |
2411 | tl = (unsigned short*)((char*)tl + len); | |
2412 | if (nul_terminated) | |
2413 | *((char*)tl - 1) = 0; | |
2414 | return tl; | |
2415 | } | |
2416 | ||
2417 | static unsigned short * | |
2418 | tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len) | |
2419 | { | |
2420 | return __tl_add_blob(tl, tag, data, len, 0); | |
2421 | } | |
2422 | ||
2423 | static unsigned short * | |
2424 | tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str) | |
2425 | { | |
2426 | return __tl_add_blob(tl, tag, str, strlen(str)+1, 0); | |
2427 | } | |
2428 | ||
2429 | static unsigned short * | |
2430 | tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val) | |
2431 | { | |
2432 | put_unaligned(tag, tl++); | |
2433 | switch(tag_type(tag)) { | |
2434 | case TT_INTEGER: | |
2435 | put_unaligned(sizeof(int), tl++); | |
2436 | put_unaligned(*(int *)val, (int *)tl); | |
2437 | tl = (unsigned short*)((char*)tl+sizeof(int)); | |
2438 | break; | |
2439 | case TT_INT64: | |
2440 | put_unaligned(sizeof(u64), tl++); | |
2441 | put_unaligned(*(u64 *)val, (u64 *)tl); | |
2442 | tl = (unsigned short*)((char*)tl+sizeof(u64)); | |
2443 | break; | |
2444 | default: | |
2445 | /* someone did something stupid. */ | |
2446 | ; | |
2447 | } | |
2448 | return tl; | |
2449 | } | |
2450 | ||
2451 | void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state) | |
2452 | { | |
2453 | char buffer[sizeof(struct cn_msg)+ | |
2454 | sizeof(struct drbd_nl_cfg_reply)+ | |
2455 | sizeof(struct get_state_tag_len_struct)+ | |
2456 | sizeof(short int)]; | |
2457 | struct cn_msg *cn_reply = (struct cn_msg *) buffer; | |
2458 | struct drbd_nl_cfg_reply *reply = | |
2459 | (struct drbd_nl_cfg_reply *)cn_reply->data; | |
2460 | unsigned short *tl = reply->tag_list; | |
2461 | ||
2462 | /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ | |
2463 | ||
2464 | tl = get_state_to_tags(mdev, (struct get_state *)&state, tl); | |
2465 | ||
2466 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2467 | ||
2468 | cn_reply->id.idx = CN_IDX_DRBD; | |
2469 | cn_reply->id.val = CN_VAL_DRBD; | |
2470 | ||
2471 | cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); | |
2472 | cn_reply->ack = 0; /* not used here. */ | |
2473 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + | |
2474 | (int)((char *)tl - (char *)reply->tag_list); | |
2475 | cn_reply->flags = 0; | |
2476 | ||
2477 | reply->packet_type = P_get_state; | |
2478 | reply->minor = mdev_to_minor(mdev); | |
2479 | reply->ret_code = NO_ERROR; | |
2480 | ||
b411b363 PR |
2481 | cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); |
2482 | } | |
2483 | ||
2484 | void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name) | |
2485 | { | |
2486 | char buffer[sizeof(struct cn_msg)+ | |
2487 | sizeof(struct drbd_nl_cfg_reply)+ | |
2488 | sizeof(struct call_helper_tag_len_struct)+ | |
2489 | sizeof(short int)]; | |
2490 | struct cn_msg *cn_reply = (struct cn_msg *) buffer; | |
2491 | struct drbd_nl_cfg_reply *reply = | |
2492 | (struct drbd_nl_cfg_reply *)cn_reply->data; | |
2493 | unsigned short *tl = reply->tag_list; | |
2494 | ||
2495 | /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ | |
2496 | ||
2497 | tl = tl_add_str(tl, T_helper, helper_name); | |
2498 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2499 | ||
2500 | cn_reply->id.idx = CN_IDX_DRBD; | |
2501 | cn_reply->id.val = CN_VAL_DRBD; | |
2502 | ||
2503 | cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); | |
2504 | cn_reply->ack = 0; /* not used here. */ | |
2505 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + | |
2506 | (int)((char *)tl - (char *)reply->tag_list); | |
2507 | cn_reply->flags = 0; | |
2508 | ||
2509 | reply->packet_type = P_call_helper; | |
2510 | reply->minor = mdev_to_minor(mdev); | |
2511 | reply->ret_code = NO_ERROR; | |
2512 | ||
b411b363 PR |
2513 | cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); |
2514 | } | |
2515 | ||
2516 | void drbd_bcast_ee(struct drbd_conf *mdev, | |
2517 | const char *reason, const int dgs, | |
2518 | const char* seen_hash, const char* calc_hash, | |
2519 | const struct drbd_epoch_entry* e) | |
2520 | { | |
2521 | struct cn_msg *cn_reply; | |
2522 | struct drbd_nl_cfg_reply *reply; | |
b411b363 | 2523 | unsigned short *tl; |
45bb912b LE |
2524 | struct page *page; |
2525 | unsigned len; | |
b411b363 PR |
2526 | |
2527 | if (!e) | |
2528 | return; | |
2529 | if (!reason || !reason[0]) | |
2530 | return; | |
2531 | ||
2532 | /* apparently we have to memcpy twice, first to prepare the data for the | |
2533 | * struct cn_msg, then within cn_netlink_send from the cn_msg to the | |
2534 | * netlink skb. */ | |
2535 | /* receiver thread context, which is not in the writeout path (of this node), | |
2536 | * but may be in the writeout path of the _other_ node. | |
2537 | * GFP_NOIO to avoid potential "distributed deadlock". */ | |
3e3a7766 | 2538 | cn_reply = kzalloc( |
b411b363 PR |
2539 | sizeof(struct cn_msg)+ |
2540 | sizeof(struct drbd_nl_cfg_reply)+ | |
2541 | sizeof(struct dump_ee_tag_len_struct)+ | |
2542 | sizeof(short int), | |
2543 | GFP_NOIO); | |
2544 | ||
2545 | if (!cn_reply) { | |
2546 | dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n", | |
2547 | (unsigned long long)e->sector, e->size); | |
2548 | return; | |
2549 | } | |
2550 | ||
2551 | reply = (struct drbd_nl_cfg_reply*)cn_reply->data; | |
2552 | tl = reply->tag_list; | |
2553 | ||
2554 | tl = tl_add_str(tl, T_dump_ee_reason, reason); | |
2555 | tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs); | |
2556 | tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs); | |
2557 | tl = tl_add_int(tl, T_ee_sector, &e->sector); | |
2558 | tl = tl_add_int(tl, T_ee_block_id, &e->block_id); | |
2559 | ||
3129b1b9 LE |
2560 | /* dump the first 32k */ |
2561 | len = min_t(unsigned, e->size, 32 << 10); | |
b411b363 | 2562 | put_unaligned(T_ee_data, tl++); |
3129b1b9 | 2563 | put_unaligned(len, tl++); |
b411b363 | 2564 | |
45bb912b LE |
2565 | page = e->pages; |
2566 | page_chain_for_each(page) { | |
cfd8005c | 2567 | void *d = kmap_atomic(page); |
45bb912b LE |
2568 | unsigned l = min_t(unsigned, len, PAGE_SIZE); |
2569 | memcpy(tl, d, l); | |
cfd8005c | 2570 | kunmap_atomic(d); |
45bb912b LE |
2571 | tl = (unsigned short*)((char*)tl + l); |
2572 | len -= l; | |
3129b1b9 LE |
2573 | if (len == 0) |
2574 | break; | |
b411b363 PR |
2575 | } |
2576 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2577 | ||
2578 | cn_reply->id.idx = CN_IDX_DRBD; | |
2579 | cn_reply->id.val = CN_VAL_DRBD; | |
2580 | ||
2581 | cn_reply->seq = atomic_add_return(1,&drbd_nl_seq); | |
2582 | cn_reply->ack = 0; // not used here. | |
2583 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + | |
2584 | (int)((char*)tl - (char*)reply->tag_list); | |
2585 | cn_reply->flags = 0; | |
2586 | ||
2587 | reply->packet_type = P_dump_ee; | |
2588 | reply->minor = mdev_to_minor(mdev); | |
2589 | reply->ret_code = NO_ERROR; | |
2590 | ||
b411b363 PR |
2591 | cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); |
2592 | kfree(cn_reply); | |
2593 | } | |
2594 | ||
2595 | void drbd_bcast_sync_progress(struct drbd_conf *mdev) | |
2596 | { | |
2597 | char buffer[sizeof(struct cn_msg)+ | |
2598 | sizeof(struct drbd_nl_cfg_reply)+ | |
2599 | sizeof(struct sync_progress_tag_len_struct)+ | |
2600 | sizeof(short int)]; | |
2601 | struct cn_msg *cn_reply = (struct cn_msg *) buffer; | |
2602 | struct drbd_nl_cfg_reply *reply = | |
2603 | (struct drbd_nl_cfg_reply *)cn_reply->data; | |
2604 | unsigned short *tl = reply->tag_list; | |
2605 | unsigned long rs_left; | |
2606 | unsigned int res; | |
2607 | ||
2608 | /* no local ref, no bitmap, no syncer progress, no broadcast. */ | |
2609 | if (!get_ldev(mdev)) | |
2610 | return; | |
2611 | drbd_get_syncer_progress(mdev, &rs_left, &res); | |
2612 | put_ldev(mdev); | |
2613 | ||
2614 | tl = tl_add_int(tl, T_sync_progress, &res); | |
2615 | put_unaligned(TT_END, tl++); /* Close the tag list */ | |
2616 | ||
2617 | cn_reply->id.idx = CN_IDX_DRBD; | |
2618 | cn_reply->id.val = CN_VAL_DRBD; | |
2619 | ||
2620 | cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); | |
2621 | cn_reply->ack = 0; /* not used here. */ | |
2622 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + | |
2623 | (int)((char *)tl - (char *)reply->tag_list); | |
2624 | cn_reply->flags = 0; | |
2625 | ||
2626 | reply->packet_type = P_sync_progress; | |
2627 | reply->minor = mdev_to_minor(mdev); | |
2628 | reply->ret_code = NO_ERROR; | |
2629 | ||
b411b363 PR |
2630 | cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); |
2631 | } | |
2632 | ||
2633 | int __init drbd_nl_init(void) | |
2634 | { | |
2635 | static struct cb_id cn_id_drbd; | |
2636 | int err, try=10; | |
2637 | ||
2638 | cn_id_drbd.val = CN_VAL_DRBD; | |
2639 | do { | |
2640 | cn_id_drbd.idx = cn_idx; | |
2641 | err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback); | |
2642 | if (!err) | |
2643 | break; | |
2644 | cn_idx = (cn_idx + CN_IDX_STEP); | |
2645 | } while (try--); | |
2646 | ||
2647 | if (err) { | |
2648 | printk(KERN_ERR "drbd: cn_drbd failed to register\n"); | |
2649 | return err; | |
2650 | } | |
2651 | ||
2652 | return 0; | |
2653 | } | |
2654 | ||
2655 | void drbd_nl_cleanup(void) | |
2656 | { | |
2657 | static struct cb_id cn_id_drbd; | |
2658 | ||
2659 | cn_id_drbd.idx = cn_idx; | |
2660 | cn_id_drbd.val = CN_VAL_DRBD; | |
2661 | ||
2662 | cn_del_callback(&cn_id_drbd); | |
2663 | } | |
2664 | ||
2665 | void drbd_nl_send_reply(struct cn_msg *req, int ret_code) | |
2666 | { | |
2667 | char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)]; | |
2668 | struct cn_msg *cn_reply = (struct cn_msg *) buffer; | |
2669 | struct drbd_nl_cfg_reply *reply = | |
2670 | (struct drbd_nl_cfg_reply *)cn_reply->data; | |
2671 | int rr; | |
2672 | ||
3e3a7766 | 2673 | memset(buffer, 0, sizeof(buffer)); |
b411b363 PR |
2674 | cn_reply->id = req->id; |
2675 | ||
2676 | cn_reply->seq = req->seq; | |
2677 | cn_reply->ack = req->ack + 1; | |
2678 | cn_reply->len = sizeof(struct drbd_nl_cfg_reply); | |
2679 | cn_reply->flags = 0; | |
2680 | ||
42ff269d | 2681 | reply->packet_type = P_return_code_only; |
b411b363 PR |
2682 | reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor; |
2683 | reply->ret_code = ret_code; | |
2684 | ||
b411b363 PR |
2685 | rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); |
2686 | if (rr && rr != -ESRCH) | |
2687 | printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); | |
2688 | } | |
2689 |